Power series 15

DIGITAL
PROTECTION
FOR
POWER
SYSTEMS
A. T. Johns and
S. K. Salman

Peter Peregrinus Ltd. on behalf of

The Institution of Electrical Engineers
IEE POWER SERIES 15

Series Editors: Professor A. T. Johns

J. R. Platts

DIGITAL
PROTECTION
FOR
POWER
SYSTEMS
Other volumes in this series:
Volume 1 Power circuits breaker theory and design C. H. Flurscheim
(Editor)
Volume 2 Electric fuses A. Wright and P. G. Newbery
Volume 3 Z-transform electromagnetic transient analysis in high-voltage
networks W. Derek Humpage
Volume 4 Industrial microwave heating A. C. Metaxas and R. J. Meredith
Volume 5 Power system economics T. W. Berrie
Volume 6 High voltage direct current transmission J. Arrillaga
Volume 7 Insulators for high voltages J. S. T. Looms
Volume 8 Variable frequency AC motor drive systems D. Finney
Volume 9 Electricity distribution network design E. Lakervi and
E. J. Holmes
Volume 10 SF6 switchgear H. M. Ryan and G. R. Jones
Volume 11 Conduction and induction heating E. J. Davies
Volume 12 Overvoltage protection of low-voltage systems P. Hasse
Volume 13 Statistical techniques for high-voltage engineering W. Hauschild and
W. Mosch
Volume 14 Uninterruptible power supplies J. D. St. Aubyn and J. Platts (Editors)
DIGITAL
PROTECTION
FOR
POWER
SYSTEMS

A. T. Johns and
s. K. Salman

Peter Peregrinus Ltd. on behalf of

The Institution of Electrical Engineers
Published by: Peter Peregrinus Ltd., on behalf of the
Institution of Electrical Engineers, London, United Kingdom

• 1995: Peter Peregrinus Ltd.

This publication is copyright under the Berne Convention and the

Universal Copyright Convention. All rights reserved. Apart from any fair
dealing for the purposes of research or private study, or criticism or
review, as permitted under the Copyright, Designs and Patents Act, 1988,
this publication may be reproduced, stored or transmitted, in any forms or
by any means, only with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms
of licences issued by the Copyright Licensing Agency. Inquiries
concerning reproduction outside those terms should be sent to the
publishers at the undermentioned address:

Peter Peregrinus Ltd.,

The Institution of Electrical Engineers,
Michael Faraday House,
Six Hills Way, Stevenage,
Herts. SG1 2AY, United Kingdom

While the authors and the publishers believe that the information and
guidance given in this work is correct, all parties must rely upon their own
skill and judgment when making use of it. Neither the authors nor the
publishers assume any liability to anyone for any loss or damage caused
by any error or omission in the work, whether such error or omission is
the result of negligence or any other cause. Any and all such liability is
disclaimed.

The moral right of the authors to be identified as authors of this work has
been asserted by them in accordance with the Copyright, Designs and
Patents Act 1988.

British Library Cataloguing in Publication Data

A CIP catalogue record for this book

is available from the British Library

ISBN 0 86341 303 X

Reprinted in paperback 1997

Printed in England by Redwood Books, Trowbridge

 Contents

Page
1 Introduction 1
1.1 Historical background 1
1.2 Performance and operational characteristics of digital
protection 2
1.2.1 Reliability 2
1.2.2 Flexibility 3
1.2.3 Operational performance 3
1.2.4 Cost/benefit considerations 3
1.2.5 Other features and functions 4
1.3 Basic structure of digital relays 4
1.4 References 5

2 Mathematical background to protection algorithms 7

2.1 Introduction 7
2.2 Finite difference techniques 7
2.3 Interpolation formulas 12
2.3.1 Forward interpolation 12
2.3.2 Backward interpolation 14
2.3.3 Central difference interpolation 15
2.4 Numerical differentiation 16
2.5 Curve fitting and smoothing 18
2.5.1 Least-squares method 18
2.5.2 Smoothing 22
2.6 Fourier analysis 24
2.6.1 The Fourier series 24
2.6.2 The Fourier transform 28
2.7 Walsh function analysis 30
2.7.1 Definition of Walsh functions 31
2.7.2 Some fundamental properties of Walsh functions 31
2.7.3 Discrete representation of Walsh functions 33
2.7.4 The Walsh series 34
2.8 Relationship between Fourier and Walsh coefficients 35
2.9 References 38

3 Basic elements of digital protection 39

3.1 Introduction 39
3.2 Basic components of a digital relay 39
vi Contents

3.3 Signal conditioning subsystems 39

3.3.1 Transducers 39
3.3.2 Surge protection circuits 41
3.3.3 Analogue filtering 41
3.3.4 Analogue multiplexers 43
3.4 Conversion subsystem 44
3.4.1 The sampling theorem 44
3.4.2 Signal aliasing error 46
3.4.3 Sample and hold circuit 47
3.4.4 Digital multiplexing 49
3.4.5 Digital-to-analogue conversion 50
3.4.6 Analogue-to-digital conversion 52
3.5 Digital relay subsystem 56
3.6 The digital relay as a unit 57
3.7 References 59

4 Sinusoidal-wave-based algorithms 60
4.1 Introduction 60
4.2 Sample and first-derivative method 60
4.2.1 Basic formulation 60
4.2.2 Calculation of an approximation to the signal
derivatives 62
4.2.3 Error analysis 62
4.2.4 Practical considerations 64
4.3 First- and second-derivative method 65
4.3.1 Mathematical formulation 65
4.4 Two-sample technique 67
4.4.1 Prediction of values of peak (or magnitude) of signal
waveforms 67
4.4.2 Determination of phase angle between waveforms 68
4.5 Three-sample technique 68
4.6 An early relaying scheme 69
4.7 References 70

5 Fourier analysis and Walsh function based techniques 72

5.1 Introduction 72
5.2 Fourier-analysis-based algorithms 72
5.2.1 The full cycle window algorithm 73
5.2.2 Fractional-cycle window algorithms 75
5.2.3 Fourier-transform-based algorithm 79
5.3 Walsh-function-based algorithms 84
5.3.1 Basic principles 84
5.3.2 Development of basic algorithm 84
5.3.3 Algorithm for Walsh function determination 85
 Contents vii

5.3.4 Estimation of the amplitude and phase angle of

fundamental components 88
5.3.5 Determination of Walsh coefficients for pure sinusoidal
waveforms 91
5.4 References 91
6 Least squares based methods 93
6.1 Introduction 93
6.2 Integral LSQ fit 93
6.2.1 Basic assumptions 93
6.2.2 Determination of unknown coefficients 93
6.2.3 Implementation of the algorithm 96
6.3 Power series LSQ fit 96
6.3.1 Basic assumptions 96
6.3.2 Shifted waveform 96
6.3.3 Approximating the shifted waveform by a power series 97
6.4 Multi-variable series LSQ technique 100
6.4.1 Basic assumptions 100
6.4.2 Derivation of the multi-variable series 100
6.5 Determination of measured impedance estimates 102
6.6 References 102

7 Differential equation based techniques 103

7.1 Introduction 103
7.2 Representation of transmission lines with capacitance
neglected 103
7.2.1 Single-phase to ground fault 105
7.2.2 Phase-to-phase and three-phase faults 106
7.3 Differential equation protection with selected limits 107
7.3.1 Basic principles 107
7.3.2 Digital harmonic filtering by selected limits 107
7.3.3 Graphical interpretation of digital filtering by
integration over selected limits 109
7.3.4 Filtering of multiple harmonic components 109
7.4 Simultaneous differential equation techniques 110
7.4.1 Lumped series impedance based algorithms 110
7.4.2 Single PI section transmission line model based
algorithms 112
7.4.3 Development of the algorithm and basic assumptions 112
7.5 References 114
8 Fundamentals of travelling-wave based protection 115
8.1 Introduction 115
8.2 The transmission line as a distributed component 115
8.2.1 Travelling waves in assumed lossless single-phase lines 119
8.2.2 Three-phase transposed lines 123
viii Contents

8.3 Superimposed quantities and their properties 126

8.3.1 Polarity of superimposed quantities versus fault location 127
8.3.2 Interrelation between the superimposed voltage and
current quantities versus fault location 129
8.3.3 Behaviour of relaying signals at the relay and fault
locations 136
8.3.4 Superimposed component elliptical trajectories 136
8.4 Bergeron's equations 140
8.4.1 Single-phase lines 140
8.4.2 Three-phase lines 142
8.5 Discriminant functions 143
8.5.1 Single-phase lines 143
8.5.2 Three-phase lines 146
8.6 References 146

9 Travelling-wave protective schemes 148

9.1 Introduction 148
9.2 Bergeron's-equation based scheme 148
9.2.1 Principles of internal fault detection 148
9.3 Ultra-high-speed polarity comparison scheme 152
9.3.1 Basic operating principle 152
9.3.2 Description of typical implementation 152
9.4 Ultra-high-speed wave differential scheme 154
9.4.1 Operating principles 154
9.4.2 Basic description of the scheme 155
9.4.3 Digital implementation of wave differential scheme 160
9.4.4 General description of the digital relay 161
9.5 Discriminant function based scheme 164
9.5.1 Operating principles 164
9.6 Superimposed component trajectory based scheme 164
9.6.1 Basic principles 164
9.6.2 Sense of trajectories versus fault direction 166
9.6.3 Extension of trajectories approach to signals including
travelling-wave components 166
9.7 References 170

10 Digital differential protection of transformers 171

10.1 Introduction 171
10.2 Principles of transformer protection 171
10.2.1 Basic principles 171
10.2.2 Biased differential relaying 172
10.2.3 Harmonic-restrained differential relay 172
10.3 Finite-duration impulse response filter based algorithms 174
10.3.1 FIR filter characteristics 174
 Contents ix

10.3.2 Extraction of fundamental and second-harmonic

components 175
10.3.3 Discrimination between inrush and internal fault
currents 178
10.4 Least-squares curve fitting based algorithms 179
10.4.1 Basic assumptions and algorithm derivation 179
10.4.2 Basis of discrimination between inrush and internal
fault currents 180
10.5 Fourier-based algorithm 181
10.5.1 Filtering of harmonics 181
10.6 Flux-restrained current differential relay 183
10.6.1 Development of the algorithm 184
10.7 Basic hardware of microprocessor-based transformer
protection 187
10.8 References 189

11 Digital line differential protection 190

11.1 Introduction 190
11.2 Current-based differential schemes 191
11.2.1 Basic principles of line current differential protection 191
11.2.2 Frequency modulation current differential
protective scheme 191
11.2.3 Modal current based protection scheme 193
11.3 Composite voltage-and current-based scheme 196
11.3.1 Basic operating principles 196
11.3.2 Formation of terminal signals 197
11.4 References 199

Index 200
 Chapter 1
Introduction

1.1 Historical background

Digital-based relaying was first contemplated during the late 1960s. In one of
the earliest papers on the subject, Rockefeller [1] suggested that all the power
system equipment in a substation could be protected using digital computers,
and since that time, research in digital protection has attracted many investi-
gators. Research activity has covered virtually every protection technique, and
many novel algorithms and associated hardware implementations have
emerged.
In 1971, Mann and Morrison [2] suggested a distance protection algorithm
based on the prediction of the peak values of current and voltage waveforms
using a sample value and its derivative. The key assumption of this approach is
that the current and voltage waveforms are sinusoidal. In 1972 a similar
technique, based on the first and second derivative, was developed jointly by
Westinghouse and Pacific Gas & Electric Company in the USA [3, 4]. Based on
this technique an experimental online system for the distance protection of a
transmission line was installed in a 230 kV substation for field trial. In 1975
Makino and Miki [5] suggested using two samples to predict the peak values of
the fault waveforms while Gilbert and Shovlin [6] developed an algorithm for
the prediction of the peak values using three samples. In both cases the authors
again assumed that the faulted current and voltage waveforms are pure
sinusoids.
Other developments, which occurred at about the same time, took into
account the non-sinusoidal nature of the faulted current and voltage waveforms.
The investigators concentrated their efforts on extracting the fundamental
components from corrupted waveforms. One approach to achieve this is based
on using Fourier methods. In 1971 Ramamoorty [7] suggested extracting the
desired fundamental component of voltage or current from faulted waveforms
by correlating one cycle of data samples with stored samples of reference
fundamental sine and cosine waves. Other workers proposed using Fourier
series to determine current and voltage fundamentals [8].
Johns and Martin [9] were the first to apply the Fourier transform to both
current and voltage waveforms using a data window of less than one cycle of
power frequency, and current and voltage transforms were then used to
calculate the measured impedance. Walsh functions were also suggested by
Horton in 1975 [10]. Girgis and Brown [11] then proposed the use of Kalman
2 Introduction

filters to extract fundamental frequency components from noisy waveforms and,

in 1982, this work was extended to the design of a digital distance protection
[12].
Many proposed methods have emerged that make no particular assumptions
as to the form of the faulted waveforms, and a number of algorithms were
developed that are based on representing transmission lines by either first- or
second-order differential equations. In 1970, Mclnnes and Morrison [13]
proposed that transmission lines be modelled as a series RL circuit, which
resulted in a first-order differential equation. This method was improved in
1975 by Ranjbar and Cory [14] by integrating the differential equations, to
determine the resistance and the inductance of the line in such a way as to
eliminate any particular harmonic and its multiples. The idea was then
developed for an experimental computer relaying system by General Electric
Co. in 1979 [15, 16]. In the same year Smolinski [17] proposed an algorithm
which included a capacitive element within the series RL model of the line. This
resulted in a second-order differential equation, which can be used to determine
the line resistance R and inductance L using four samples.
The introduction of UHV lines in particular brought about new and
relatively difficult problems, which required new approaches to line protection.
The new methods developed were based on what are often called travelling-
wave techniques. The use of such techniques has been reported since 1977 by a
number of investigators, notably Tagaki et al [18, 19], Dommel and Michels
[20], Chamia and Liberman [21], Yee and Esztergalyos [22], Johns [23, 24],
Vitins [25], Crossley and MacLaren [26] and Mansour and Swift [27, 28]. The
intense activity in this field has resulted in several commercial developments,
particularly in the field of directional comparison protection.

1.2 Performance and operational characteristics of digital

protection
It has been recognised that many benefits can be gained from the application of
digital protection and these can be broadly classified under five main areas.

1.2.1 Reliability
Digital relays can be designed to regularly monitor themselves. The process of
monitoring involves executing the relay software in conjunction with a
prespecified data set and comparing the results with those expected from a
properly functioning device. If the response turns out to be different from that
expected, an error is detected and the relay initiates warning signals to the
operator. This feature can be extended by programming the relay to monitor its
peripherals. It should be noted that self monitoring does not in itself directly
improve reliability, but it does provide a means of signifying the operational
state of protection equipment. This in turn has an indirect beneficial effect on
 Introduction 3

overall reliability by ensuring that the number of potential malfunctions is

reduced. Reliability can be improved further by building a degree of redun-
dancy into the hardware/software design and using different operating princi-
ples within the same relay.

1.2.2 Flexibility

Digital relays are generally more flexible than conventional devices. For
example, digital relays are programmable, and this in turn makes it possible to
use the same hardware for performing a variety of protection and control
functions by effecting changes in the software. It is also possible that the same
relay can be equipped with multiple characteristics and any revisions or
modifications required by changes in the operational conditions of the system
can be easily accommodated with virtually no changes in the hardware
structure.

1.23 Operational performance

Research and field experiments have shown that, in difficult applications in

particular, digital relays can be arranged to perform much better than
conventional relays. This is particularly so in long distance EHV/UHV
transmission lines, series/shunt-compensated lines and multi-ended circuits. It
is also recognised that certain features are naturally inherent to digital relays,
e.g. memory action and complex shaping of operational characteristics.

1.2.4 Cost/benefit considerations

The cost of conventional relays has continued to increase during the last two
decades and the cost/benefit ratio has consequently generally increased. On the
other hand, the advancement in microelectronic technology has led to a
substantial reduction in the cost of digital hardware but it must be remembered
that, in particular, it is the cost of the software that often dominates the overall
cost. Situations exist where the cost of software for commercially developed
equipment exceeds that of the hardware by at least an order of magnitude, and
in consequence digitally based equipment costs more than conventional
equipment.
On the other hand, high-volume digital relays, e.g. overcurrent relays, are
relatively cheap because the development costs are spread across many relays
and volume production allows the use of special microchip technology. Sales
volumes and development costs are also important; these vary significantly
according to the degree of functional complexity involved. Overall, it is true to
say that the substantial improvements in performance made possible by the
application of digital technology have resulted in a gradual reduction in the
cost/benefit ratio for digital-protection equipment.
4 Introduction

1.2.5 Other features and functions

With the introduction of microprocessor-based protective systems, totally new
features and facilities, which have no parallel in conventional technology, have
been made possible. In particular, digital relays can be programmed to provide
post-fault analysis of all observed transient phenomena. This is achieved by
reading out sampled data that have otherwise been acquired as part of the fault-
measurement process. In addition, digital equipment monitoring both voltage
and current can be programmed to compute the distance to a fault immediately
after the occurrence of the fault. Such information is often extremely useful to
maintenance teams in the inspection of lines following sustained fault clearance.
Improved fault-location techniques are assuming major importance, particu-
larly in long transmission lines and those subject to a relatively high incidence
of contingency faults over difficult terrain. The general reduction in fault-
clearance times is resulting in a reduction in visual evidence of faults, and this in
turn is further fuelling a demand for accurate fault location facilities as an
integral part of digital protection devices.

1.3 Basic structure of digital relays

Unlike conventional relays, a digital relay consists of two main parts: hardware
and software. The type of software embedded in a relay decides not only its
characteristics but its function as well, i.e. whether it is an overcurrent,
differential or impedance-based measuring device. An integral and important
part of the software is the algorithm, which is a set of mathematical instructions
used to process input currents and/or voltages to estimate system parameters
such as the RMS values of signal components, measured impedance, funda-
mental frequency and differential currents etc. These calculated parameters are
then used to decide whether the system is sound or faulty, and consequently
initiate the action necessary to isolate the faulted section.
Over the past 20 years, different types of algorithms have been developed for
digital protection applications and these may be broadly classified:
(i) Sinusoidal waveform-based algorithms, which include algorithms such as
sample and first derivatives, first and second derivatives, two-sample
techniques and three-sample techniques.
(ii) Fourier and Walsh-based techniques.
(iii) Least-square methods.
(iv) Solution of the differential equations of a protected-system model.
(v) Travelling-wave-based methods.
Much of the work presented in this book will be concerned with the general
principles underlying the development and implementation of the above
detailed algorithms. They can be applied to the protection of generator
transformers, lines, busbars, switchgear and cable circuits in both transmission
and distribution systems. The range of prospective applications is vast, but the
 Introduction 5

main applications to date have been in the areas of lines and transformers in
transmission systems. This book will concentrate on the latter applications, but
without reference to specific commercial implementations, details of which are
available from manufacturers' product literature.

1.4 References
1 ROCKEFELLER, G.D.: 'Fault protection with a digital computer', IEEE Trans.,
1969, PAS-88, pp. 438-461
2 MANN, B.J., and MORRISON, I.F.: 'Digital calculation of impedance for trans-
mission line protection', IEEE Trans., 1971, PAS-90, 1971, pp. 270-279
3 GILCHRIST, G.B., ROCKEFELLER, G.D., and UDREN, E.A.: 'High-speed
distance relaying using a digital computer, Part I: System description', IEEE Trans.,
PAS-91, 1972, pp. 1235-1243
4 ROCKEFELLER, G.D., and UDREN, E.A.: 'High-speed distance relaying using a
digital computer, Part II,' ibid, pp. 1244-1258
5 MAKINO, J., and MIKI, Y.: 'Study of operating principles and digital filters for
protective relays with a digital computer', IEEE Publ. No. 75 CH0990-2 PWR,
Paper C75 197 9, IEEE PES Winter Power Meeting, New York, Jan. 1975, pp. 1-8
6 GILBERT, J.G., and SHOVLIN, R.J.: 'High speed transmission line fault impe-
dance calculation using a dedicated minicomputer', IEEE Trans., 1975, PAS-94,
1975, pp. 544-550
7 RAMAMOORTY, M.: 'Application of digital computers to power system protec-
t i o n ' , / Inst. Eng. (India), 1972, 52, pp. 235-238
8 McLAREN, P.G., and REDFERN, M.A.: 'Fourier-series techniques applied to
distance protection', Proc. IEE, 1975, 122, pp. 1295-1300
9 JOHNS, A.T., and MARTIN, M.A.: 'Fundamental digital approach to the distance
protection of EHV transmission lines', Proc. IEE, 1978, 125, pp. 377-384
10 HORTON, J.W.: 'The use of Walsh functions for high-speed digital relaying', IEEE
PES Summer meeting, San Francisco, July 20-25, 1975, Paper A 75 582 7
11 GIRGIS, A.A., and BROWN, R.G.: 'Application of Kalman filtering in computer
relaying', IEEE Trans., 1981, PAS-100, pp. 3387-3397
12 GIRGIS, A.A.: 'A new Kalman filtering based digital distance relay', ibid, 1982,
PAS-101, pp. 3471-3480
13 McINNES, A.D., and MORRISON, I.F.: 'Real time calculation of resistance and
reactance for transmission line protection by digital computer', Elec. Eng. Trans. Inst.
Eng. Australia, 1970, EE7, pp. 16-23
14 RANJBAR, A.M., and CORY, B J.: 'An improved method for the digital protection
of high voltage transmission lines', IEEE Trans. 1975, PAS-94, pp. 544-550
15 BREINGAN, W.D., and CHEN, M.M.: 'The laboratory investigation of a digital
system for the protection of transmission lines', IEEE Trans. 1979, PAS-98, pp. 350-
368
16 GALLEN, T.F., and BREINGAN, W.D.: 'A digital system for directional compari-
son relay', ibid, 1979, PAS-98 pp 948-956
17 SMOLINSKI, W.J.: 'An algorithm for digital impedance calculation using a single
PI section transmission line model', ibid, 1979, PAS-98, pp. 1546-1551
18 TAKAGI, T., BABA, J., VEMURA, K., and SAKAGUCHI, T.: 'Fault protection
based on travelling wave theory, Part I: Theory', IEEE PES Summer Power meeting,
Mexico City, July 1977, Paper No A 77, pp. 750-753
19 TAKAGI, T., BABA, J., VEMURA, K., and SAKAGUCHI, T.: 'Fault protection
based on travelling wave theory, Part II: Sensitivity analysis and laboratory test',
IEEE PES Winter meeting, New York, 1978, Paper A 78, pp. 220-226
20 DOMMEL, H.W., and MICHELS, J.M.: 'High speed relaying using travelling
wave transient analysis', IEEE PES Winter meeting, New York, 1978, Paper A 78,
pp. 214-219
6 Introduction

21 CHAMIA, M , and LIBERMAN, S.: 'Ultra high speed relay for EHV/UHV
transmission lines - Development, design and application', IEEE Trans. 1978,
PAS-97, pp. 2104-2116
22 YEE, M.T., and ESZTERGALYOS, J.: 'Ultra high speed relay for EHV/UHV
transmission lines - Installation, staged fault tests and operational experience', ibid,
1978, PAS-97, pp. 1814-1825
23 JOHNS, A.T.: 'New ultra-high speed directional comparison technique for the
protection of e.h.v. transmission lines', IEE Proc, 1980, 127, pp. 228-238
24 JOHNS, A.T., MARTIN, M.A., BARKER, A., WALKER, E.P., and CROSSLEY,
P.A.: 'A new approach to e.h.v. direction comparison protection using digital signal
processing techniques', IEEE Trans., 1986, PWRD-1, pp. 24-34
25 VITINS, M.: 'A fundamental concept for high speed relay', IEEE Trans., 1981,
PAS-100, pp. 163-173
26 CROSSLEY, P.A., and McLAREN, P.G.: 'Distance protection based on travelling
waves', IEEE Trans., 1983, PAS-102, pp. 2971-2982
27 MANSOUR, M.M., and SWIFT, G.W.: 'A multi-microprocessor based travelling
wave relay - Theory and realization', IEEE Trans., 1986, PWRD-1, pp. 273-279
28 MANSOUR, M.M., and SWIFT, G.W.: 'Design and testing of a multi-
microprocessor travelling wave relay', IEEE Trans., 1986, PWRD-1, pp. 74-82
 Chapter 2
Mathematical background to protection
algorithms

2.1 Introduction

Digital protection devices involve extensive use of numerical techniques, which

are implemented in real time. Such methods are often specific to digital
protection, and as such are often alien to engineers trained in the development
and application of previous generations of analogue or electromagnetic-based
protection devices.
To understand the principles underlying digital protection technology, it is
necessary to review briefly the mathematical basis of numerical algorithms. The
topics covered in this Chapter therefore include finite differences, numerical
differentiation, curve fitting and smoothing, Fourier analysis, Walsh analysis,
and the relationship between Fourier and Walsh coefficients. It is not the
intention that the material presented should be highly rigorous in the mathema-
tical sense, but rather that it should give a working knowledge of the numerical
techniques used and thus provide a basis for the work on specific protection
algorithms that is presented in later Chapters.

2.2 Finite difference techniques [1, 2, 3]

Let us assume that numerical values f{xk) of some function y =/( x) are given at
equally spaced values of x, such that x, = #0 + A, x2 = x0 + 2h, . . . , xk = x0 + kh.
f(xk) can be obtained either analytically from some formula or by signal
sampling. Therefore at x = xk

IM=f{%+kh)=£ (2.1)
Using given numerical values fki it is possible to form difference tables
containing differences up to any desired order n.
In general there are three difference functions that can be derived for a given
sample set. These are forward differences, backward differences and central
differences. In digital protection, such differences are commonly used in the
8 Mathematical background to protection algorithms

Table 2.1 Forward difference table for a function fk = 1 /\/xk

4=1/V^ A% A 4f A 5/
x*

1 1.0000
-0.2929
2 0.7071 0.1631
-0.1298 -0.1106
3 0.5774 0.0525 0.0827
-0.0774 -0.0279 -0.0656
4 0.5000 0.0246 0.0171
-0.0528 -0.0108
5 0.4472 0.0138
-0.0390
6 0.4083

process of deriving, from sampled data, approximations to the rate of change

(or first difference) and second differences of relaying signals.
The forward difference functions are defined as follows:

It follows that, in general, for any positive value of n, the nth order difference
equation takes the form of

i+,~ A (2.2)

By way of illustration, Table 2.1 thus shows the difference Table for the
function/(*) = l/V*.
Backward differences are basically defined in terms of present and preceding
sampled values, so that the first and second backward differences are given by

Vfk =fk — fk_ | for the first backward difference

and

for the second

The nth backward difference is thus

(2.3)
 Mathematical background to protection algorithms 9

Finally, the central difference function is defined in terms of values sampled at

one half of the sample interval {hi2) on each side of the time at which an
estimate is made. Thus, the first and second central difference functions are
given by:

yfk =fk+\/2 ~~Jt-I/2

and

The nth central difference function is thus

5 5l(5"-|/,- 1 / 2 (2.4)

Since samples are not usually available at half intervals, the easiest way to
implement central difference techniques is by using the samples before and after
the current sample, i.e. **__, and xk+l. Thus, for example, the first central
difference becomes

Given a sampled data set taken at x*_{, xh xk+u . . . , the sequences of numbers
defining the forward and backward difference functions are identical. The
distinction between the three functions essentially lies in the time at which a
given function is available for use in further digital processing and, where a
difference function is used in approximating the differential of an actual signal
(to whatever order), the degree of the resulting approximation. Thus, for
example, the forward difference function is available at the time the k 4- 1 th
sample is available whereas that for the backward approximation is available at
the kih sample instant. Furthermore, the degree of accuracy to which a
particular function can be used to approximate, say, the first differential of a
signal varies according to the waveform involved. For example, with reference
to Figure 2.1, it will be evident that a central difference function using samples
taken at instants corresponding to k + 1/2, k— 1/2 may give a better approxi-
mation to the differential of the particular signal shown than is the case for the
forward or backward difference function. In other words, the slope of a straight
line joining BD in Figure 2.1, which corresponds to the first derivative
calculated using the central difference formula, is close to the slope of the
tangent at C.
In practice, it is not generally possible to assume a particular form to the
waveforms encountered and the accuracy in approximation that can be
achieved with each method varies from point to point as a given waveform is
tracked. Each method therefore has advantages in particular applications. In
any event, the basic difference functions form the basis of more accurate
approximations to signal derivations by using interpolation formulas explained
in Section 2.4.
10 Mathematical background to protection algorithms

Figure 2.1 Forward, backward and centre difference function related to an arbitrary
waveform

Closely associated with difference function operators A, V and d are the shift
operator E and the average operator fi. The shift operator is defined as the
operator that increases the instant at which a function is sampled by one
tabular interval. Thus

E(Jk)=f(Xt+h) (2.5)

A relation between the shift operator and finite difference operators can be
derived as follows:
 Mathematical background to protection algorithms 11

From the definition of A/J, we have

4 / * = i + , -I = Efk -fk =(E- \)fk

Therefore

A= £~l (2.6)

or

£=1+A (2.7)

Also from the definition of backward difference operator (V) we obtain

V/, = (£-l)i_, (2.8)

from which:

V=l-£-! (2.9)

The first central difference equation

=Z
vfk fk+ 1/2 ~~Jt-l/2

can now be written in terms of the shift operator:

6fk = E[nfk-E-Uifk = (Em -E~m )fk (2.10)
It is thus apparent from eqn. 2.10 that the centre difference operator can be
written in terms of the shift operator:

6 = Em-E'm (2.11)
The average operator ju is defined as follows:

^=l/2(/*+l/2+/,-1/2) (2.12)

In effect, the average operator permits an estimate of a given function to be

obtained from sampled values taken at half-sample intervals on each side of the
time at which an estimate is required.
Using the definition of the operator E, the equation thus becomes:

or
m
+ E-m) (2.13)
12 Mathematical background to protection algorithms

The various operators detailed above represent a succinct way of representing

sampled values. They also, importantly, allow easy manipulation of the
sampled data used in algorithmic equations and are consequently widely used
in the development of software for digital protection systems.

2.3 Interpolation formulas [1, 2, 3]

Different methods can be used to derive interpolating formulas. A very common

approach, which is used here, is based on the previously explained shift
operator.

23.1 Forward interpolation

Given a function such as that shown in Figure 2.1, its forward value
fk+i =f(xk + h) can be found from its current value^ =f(xk) using eqn. 2.5, such
that

fk+l=f(xt+h)

Similarly, the value of the function at x = xk+ph is equal to its value at xk shifted
by an interval equal to p times the sampling interval h, so that

(2.14)

Substituting the value of E from eqn. 2.7 into eqn. 2.14 we obtain

. . . +k/,fA"]fk (2.15)
where any value kmj is often known as the mth binomial coefficient, given by
p(p-\)(p~Z)...(p-m+\)
J
ml

It follows that an estimate of the value of a function can be obtained from the
summation of terms given in eqn 2.16:

r i
k
31

n\
where n is the order of the polynomial.
 Mathematical background to protection algorithms 13

Ifp is equal to rsuch that O ^ r ^ 1, t\icnfk+p=fk+r is simply equal to the value

of the function at xk + rh, and it follows that

r(r-l) r(r-l)(r-2)
ft+r =/(** + rh) =£ + rAfk + —^—I2fk ^ ^ &/>+•••

r(r-l)(r-2) . . . (r-«+l)
+- -, -*% (2.17)

Eqn. 2.17 is known as the Gregory-Newton forward interpolation formula,

which is useful in determining the value of a function from its forward difference
table.

Example
Calculate the value of the function f(x) = 1/V# at x = 3.5 from its forward
difference Table 2.1, using eqn. 2.17.

Solution
Let us first estimate the value using only the first two terms of eqn. 2.17 so that
we have^ +r —fk + rAjf, where £ = 3 and r=0.5.
With reference to Table 2.1, / 3 and A/3 are equal to 0.5774 and -0.0774,
respectively, so that

f3m5 - 0.5774 + 0.5( -0.0774) = 0.5387

This corresponds to an error of 0.786% from the exact value/(3.5) =

which equals 0.5345. Now let us repeat the calculations using the first three
terms of eqn. 2.17, which gives

r(r-l)
— ^ — A 2 /,

Again, from Table 2.1, A% = 0.0246, so

0.5(0.5-1)
/3.5 = 0.5774 + 0.5(-0.0774) + (0.0246)

= 0.535625

This corresponds to an error of 0.21%, which indicates that the more terms of
eqn. 2.17 are used, the better the accuracy obtained. However, for this example,
it is sufficient to take only the first two terms. In digital protection applications,
eqn. 2.17 can be very useful in determining a numerical estimate of the
differential of a function; this will be discussed in Section 2.4.
14 Mathematical background to protection algorithms

2.3.2 Backward interpolation

Consider the function shown in Figure 2.1; a backward value of this function,
say/_, =f(xk — h), can be used to obtain its current value/=/(%) by using the
shift operator E. Thus/_, =f(xk- A) = E~lf(xk), o r / , = £/*„, .
In general the current value/ can be found from the function/^ which isp
intervals behind it, which can be succinctly stated mathematically by / =

It is often more convenient to rename the function value fk^p as fN and

therefore the current v a l u e / becomes fN+p. Thus

(2.18)

Substituting eqn. 2.9 into eqn. 2.18, we obtain

*2*V 2 -AJHV 3 +- • -+kpbVP]fN

The binomial coefficients klh k2(, . . . take the form

/>G&+l)(/> + 2) •• • (p + m-1)
ml

It follows that the value of a function can be estimated by eqn. 2.19 by using the
backward difference operator

As previously, eqn. 2.19 can be written in terms of a function of the shift interval
h so that, for a fractional shift sh say, eqn. 2.19 becomes:

, s(s+\)(s
v 2
/+

s(s+\)(s + 2) . . . (s + n-\)
+~ ~{ % v (2.20)

This is known as the Gregory—Newton backward interpolation formula. Again,

as we shall see later in this Chapter, this formula is useful in determining an
estimate of the differential of a function.
 Mathematical background to protection algorithms 15

2.3.3 Central difference interpolation

There are a number of different versions of interpolation formulas that can be

obtained by using central differences [2]. Four of the more common formulas
are given here for reference and completeness:

(i) Forward Newton-Gauss interpolation formula

r(r-l) (r+i)(r-i)
2 3

[ (r+l)r(r-l)(r-2)j4

(ii) Backward Newton-Gauss interpolation formula

.r^r , r ( r - l ) , 2 r , (r-H)r(r-1)

(2.22)

(iii) Stirling interpolation formula

(2 23)
'

(iv) Laplace-Evcrctt interpolation formula

(2.24)
16 Mathematical background to protection algorithms

2.4 Numerical differentiation [3, 4, 5]

Any of the interpolation formulas derived in the previous Section can be used to
estimate the derivative of a sampled function. For example, let us consider the
forward interpolation formula of eqn. 2.17.
r(r-l) r(r-l)(r-2)

By differentiating with respect to r, we obtain

2r-l 3r2-6r+2
hf'(xk + rh) = Afk +—A f k + 2
A3/,
2r 3 -9r 2 +llr~3
+
12
and again
6r 2 -18r+ll
A4/,
Derivatives at the point (xh fk) can be obtained by setting r = 0 so that for the
first derivative we obtain

^ - ^ A 2 + ^ A 3 - - • ^fk (2.25)

The second derivative formula will likewise be

./!'=/"(**) =U A2 - A3 +|^ A4 - • • \fk (2.26)

Similarly, from the backward interpolation eqn. 2.20 by differentiating with

respect to s, we obtain
25-h 1 3/4-65 + 2

VfN+
Setting s = 0 leads to the first derivative, which becomes
V
/*'=/'(%=* > 4 ( +\^+l^ + - • \ft (2.27)
and

( ^ ) (2.28)
 Mathematical background to protection algorithms 17

Derivatives can alternatively be calculated from central differences using any of

the equations from 2.21 to 2.24; a useful central difference form is given in eqn.
2.29:

6 30 / ( 2 2 9 )

Example
Use the backward interpolating formula to derive an expression for estimating
the derivative of a sampled function after
(i) two samples are made available
(ii) three samples are made available, but the derivative is to be calculated for
the instant in time that the second sample is available
(iii) repeat (ii) for the second derivative.
Solution
(i) By considering only the first term of eqn. 2.27, we obtain an estimate based
on only two samples; this is given in eqn. 2.30.
1 1
fk=-^fk=-hUk-fk-x ) (2.30)
(ii) Since the derivative is required to be found following the second sample,
let us assume that the last sample is^+,, and hence the second sample
would b e ^ . Now eqn. 2.27 can be expressed in terms offk+l by using the
relationship fk = E~lfk+l which, from eqn. 2.9, can be written in the form
given in eqn. 2.31:
^ = (1-V)j£ +1 (2.31)
By substituting eqn. 2.31 into eqn. 2.27, and simplifying we obtain

(2.32)

Now by considering only the first and second terms, eqn. 2.32 reduces to
eqn. 2.33, which in turn defines, as required, an estimate of the rate of
change of the function at an instant in time midway between the pair of
samples used.

/;=^a + .-i-i) (2.33)

(iii) By substituting eqn. 2.31 into eqn. 2.28 we obtain
1 / 1 1
h\ 12 12
Thus by considering only the first term we obtain
1
h2
18 Mathematical background to protection algorithms

As required, eqn. 2.34 defines a measured approximation to the second

rate of change of a function at the time when the second sample is taken.

2.5 Curve fitting and smoothing [3-5]

The process of representing a data set by a mathematical function (or

expressing a complicated mathematical expression by a simpler one) is known
as curve fitting. A widely used procedure to achieve this involves the method of
least squares, which can be used for both tabular data and functions.

2.5.1 Least squares method

Consider a set of N measured points defined by pairs of numbers:

where x{ is the value of the independent variable taken at the fth measurement
and j>; its corresponding dependent variable. Such data may be generated, for
example, by the measurement of current flowing through an electrical element
(independent variable) and its corresponding voltage across the element
(dependent variable). In digital protection, it is common to sample a relaying
voltage at discrete instants in time; in this case, the dependent variable can be
regarded as the sampled value and the independent variable the discrete value
of time at which it is sampled. Let us now determine a function u(x) to
approximate the actual function such that

y^u(xt) i = l , 2 , ...9N (2.35)

The type of function best used in the approximation process—polynomial,

exponential etc.—depends on the nature of the problem. In most protection
cases the most widely used and suitable approximating function is a polynomial
of the form

u = ao + a{x + a2x2 + • • • +amxm (m<N) (2.36)

To define this function fully, it is necessary to choose the values of the constants
OQ, al9 . . . , am such that the function best fits the given data. One criterion,
which is often regarded as giving the best fit, requires that the sum of the
squares of the displacements of the measured points (yt) from the curve defined
by eqn. 2.36 should be a minimum.
 Mathematical background to protection algorithms 19

least-square
X yC polynomial curve (u)

vertical displacement from

least-square polynomial curve fe)

Figure 2.2 Method of least squares

Figure 2.2 illustrates the basis of curve fitting using the least-squares method.
Let the vertical displacement of measured points (jy,) from the fitted curve u be
£,, in which case:
£,=.?,-«(*,-) (2.37)
The sum of the squares of errors is thus as given by eqn 2.38.
N N

(2.38)

The necessary condition for S to be a minimum is that its partial derivative with
respect to the parameters a0, a2i . . . etc. should be zero, i.e.
dS
— =0 (2.39)
dak
When eqn. 2.39 is applied to eqn. 2.38, iV equations will be formed, often called
'normal equations'. Thus
dS
-
da0
N
dS
iirr:.... _ . c\

da i

dS
20 Mathematical background to protection algorithms

After simplification, the resulting equations in a matrix form would be

N 2* 2^
x
t 2LJ
Xi
2Ld Xi
/*iA| a
\

(2.40)
Eqn. 2.40 represents a system of simultaneous equations which can be solved
for the unknown a0, a{, . . . . am.
Example
The following data were obtained experimentally:
-1.0 -0.1 0.2 1.0

1.0
y 1.099 0.808 1.0
Use the least square method to fit a straight line to this data.
Solution
The straight line equation is given by
u = ao + axx
The normal equations for this function are obtained directly from eqn. 2.40.
Setting m = 1, as required for a straight line approximation, the normal matrix
equation is:

» 2^ a0 2* (2.41)
ax

Since there are four data points, the limit on the summations is N=4, and this
gives the following constant values in eqn. 2.41:

*,.=-1.0-0.1+0.2+1.0 = 0.1

^ 2 =(-1.0) 2 +(-0.1) 2 +(0.2) 2

 Mathematical background to protection algorithms 21

y. = i .0 + 1.099 + 0.808 +1.0 = 3.907

xji= (-1-0) (1.0) + (-0.1) (1.099) + (0.2) (0.808) + (1.0) (1.0) = 0.0517

Substituting these values into eqn. 2.41 and solving for a0 and au we obtain

ao = 0.9773 ax = -0.0224

The straight line equation which best fits the data set is therefore given by

u = 0.9773 - 0 . 0 2 2 4 *

In the previous analysis, it has been assumed that the variabley depends only
on a single variable x. The general case is such that the variabley may depend
on more than a single variable. An example that illustrates this case is a three-
phase unbalanced system in which the phase voltage va depends on the three
associated phase currents ia, ib and ic. Therefore, in order to determine a value of
va, the currents ia, ib and ic must be measured. In this case, the voltage va is a
function va=f(ia, ib, ic). Thus each measured voltage is related to the measured
currents by a data set (ial, ibU icU va{), (ia2y ib2, ic2, va2) . . . (iaNy ibN, icN, vcN).
More generally, for a function of z variables, the data set involved
would take the form (*,„ x2U xzu . . . xzUyx), (xl2, xm x32 . . . xz2iy2), . . .
(xlN, x2N, x3N, . . . xzN,yN). The principles involved in deriving a least-squares-
error based function for extended data sets associated with multivariable
functions are similar to those outlined above for the single-variable case. Full
details of the precise methodology involved can be found in References 4 and 5.

2.5.1.1 Least squares and pseudoinverse [6—9]

The solution of linear equations having unknowns less than the number of
equations resembles a situation similar to the least-squares method. For
example, let us consider the following n linear equations having m unknowns,
where n>m:

(2.42)
0«»*i + 0*2*2 + • * ' + anmxm =yn

Writing these equations in a more succinct way results in

A X = Y (2.43)
nXm mX1 nX 1
22 Mathematical background to protection algorithms

where

#22 a
2m x2 J2
A= Y=

Xm

By premultiplying eqn. 2.43 by AT (the transpose of matrix A), we obtain

AT A X = i4T F
(2.44)
mXn nXm mX\ mXn nX\

In its expanded form, eqn. 2.44 is of the form

X\

x2

(2.45)

It should be noted that the limits of all the summations in eqn. 2.45 are from
i=\ to n. Eqn. 2.45 in effect defines the normal equations obtained by using the
least-squares method. It can also be seen from eqn. 2.44 that the matrix A1 A is a
square matrix having a dimension of m X m. Therefore, by premultiplying eqn.
2.44 by [i4Ti4]~!, we obtain the vector of unknowns as given by eqn. 2.46.

X =[Al A]~l A Y (2.46)

mX\ mXm mXn nX\

The matrix [i4Ti4] l A is known as the left pseudoinverse of A and, since eqn.
2.46 is the solution of the least-square error eqn. 2.45, it is therefore clear that
the pseudoinverse approach provides the least-squares solution.

25.2 Smoothing [3]

Data obtained from physical measurement often contain errors that can be
conveniently defined as the difference between the correct and measured values.
The correct value means the true, theoretical or average value. In practice, it is
 Mathematical background to protection algorithms 23

usually desirable to eliminate as much error as possible from the measured

data, and this can be done by fitting the data to a formula derived using the
least-squares method. This process is often known as smoothing.

Example
Let us assume that it is required to find the smoothed value for a measured
three-point data set using linear equations. Let the measured data be denoted
by (^bji)? (*2J J2) a n d (#35^3)- The process of finding the smoothed values of
the given data to fit a linear equation basically consists of the following steps:
(a) The data are fitted, using the least squares method, to the linear function of
eqn. 2.47.
y = aQ-¥a,x (2.47)
This of course involves the determination of the constants a$ and ax to give
the best fit.
(b) The smoothed value Yk can then be determined at the corresponding xk by
using eqn. 2.48.

Yk = ao + alXk (2.48)
To simplify the analysis let us:
(i) Assume that the data are taken at equal spaces in the x variable, such that
xk = kh (2.49)
where h is the interval between values.
(ii) Renumber the measured data in such a way as to extend them symmetri-
cally from negative to positive around a defined middle point. Therefore the
data set becomes (*_,,j_,), (*o,Jo)> (*i,J>i)-
From previous work (see eqn. 2.41), the measured data set can be fitted to eqn.
2.47, using the least-squares method, to obtain the following normal equations:

1=1 /= I

N N N

i= 1 i= I

where N is the number of measured data points (here N= 3).

By substituting eqn. 2.49 into eqn. 2.50 and 2.51, and changing the limits as
indicated above, the normal equations become:

(kh) (2.52)
24 Mathematical background to protection algorithms

I
2
^ (2.53)
k=-\ k=-l k=-\

From eqn. 2.52 we obtain

j ^ + j o + J ' i = 3flo +fl,(-*A +O + AA)
which gives

%=O>-i+J>o+Ji)/3 (2.54)
From eqn. 2.53 we likewise obtain

which likewise gives

al = (-j.l+jx)/2k (2.55)
The smoothed values Yk corresponding to xk = kh(k = — 1,0, 1) can be found by
substituting eqn. 2.54 and eqn. 2.55 into eqn. 2.48. Thus

Therefore
y_1 = (5j_, + 2 j 0 - j , ) / 6 (2.56)
Similarly the smoothed values YQ and Yx can be found as
Fo=O'-«+J>o+J>.)/3 (2.57)
Yl = (-j.l-¥2y0^5yl)/6 (2.58)

2.6 Fourier analysis [10, 11]

2.6.1 The Fourier series
Trigonometric form
Any periodic function/(/) can normally be represented by a Fourier series of
discrete harmonics

fit) = j + 2 a
« cos na)ot + X *. sin Wo1 (2-59)
where (O0 is the angular fundamental frequency = 2Jtf0 = 2nl T
T is the time period of the fundamental component
na)0 is the nth harmonic angular frequency
t{ is arbitrary
Given the known function f(t) (which in most practical situations is a function
that varies with time), the coefficients OQ, a,, bu . . . an, bn can be determined
from expressions of the form of eqns 2.60-2.62 [10, 11].
2 f'i+r
*o= y / ( 0 dt (2.60)
J /i
 Mathematical background to protection algorithms 25

2 f^T
f(t) cos natotdt (2.61)

2 f'.
/(*) sin na)ot dt (2.62)
J /,

Alternatively, by combining corresponding sine and cosine terms of the same

frequency, eqn. 2.59 can be written as

^ (2.63)
n=0

where

A0 = j,00 =0 (2.64)

and

"1 - (n= 1, 2, . . .) (2.65)

In some applications it is more convenient to use the complex form of Fourier

series given in eqn. 2.66. It will be apparent that the latter equation is directly
equivalent to the basic Fourier series expression given in eqn. 2.59.

2 ^ e>l °°' (2-66)

n=-oo

where
£ = ( f l . - A ) / 2 , B = ± l , ± 2 . . . . ±00

By substituting eqns. 2.60—2.62 into the above equation, Fn reduces to

_ 1 /Vr
/(/) e-jn^dt, » = ± 1, ± 2 . . . ±oo (2.67)

-I f" r
f(t)dt,n = O

The spectrum of periodic signals

It will be apparent from the previous section that any periodic function f(t) can
be represented in one of two ways:
(i) a time-domain representation, where the time history of the function
behaviour is described by using the equation which defines the function/(/)
26 Mathematical background to protection algorithms

(ii) a frequency-domain representation, where the waveform is described by

the magnitude and phase of a number of sinusoidal components of
frequencies na)0(n= 1 , 2 , . . . ) that make up the signal. This is done by
specifying all the Fourier coefficients, i.e. any one of the three sets {am bn),
{Am 8.) or Fa.

Example
Figure 2.3 shows a periodic function f(t) with a period of T, which is given by
A

mT-\ mT+- = 0, ± 1 , ± 2 , . . . ±«>

f(t) = (2.68)
. o elsewhere
Find its representation in the frequency domain.

Solution
Representation of the function f(t) in the frequency domain requires the
determination of its Fourier coefficients:
From eqn. 2.67

=-f"
+T

* TJ
f(t) , «= ± 1, ± 2, . . . ±00
t,

f(t)dt,n =

Using the definition o£f(t) given in eqn. 2.68 and taking the limits of integration
from — r/2 to r/2, we obtain
1 f'/i
dt, n = ±l, ± 2 , . . . ± o o

Ar-jna>Qt
~T[-jnco
">«>oJ_
0 r/2

f(t)

T -c/2 x/2
Figure 2.3 Illustrative periodic function
 Mathematical background to protection algorithms 27

At _ A
T "" 2

Figure 2.4 Fourier coefficients Fn (n — 0, 1, . . .) versus nfor the periodic function shown
in Fig. 2.3 for (a) z/T= 1/2, and (b) x/T= 1/4

At

At
(2.69)

where

sin x
sinc(^) =-

The value of Fn(n = Q, ± 1 , ± 2 , . . .) defines the Fourier coefficients corres-

ponding to discrete angular frequencies ajn, « = 0, 1, 2, . . . . By knowing the
ratio t/T, it is possible to determine Fn and Figure 2A(a) and (b) shows Fn
versus (ott for ratios x/T= 1/2 and t/T= 1/4 respectively. These figures also
represent the function f{t) in the frequency domain.
28 Mathematical background to protection algorithms

2.6.2 The Fourier transform

We have seen in previous Sections how any periodic function can be treated
using the Fourier series. However, in practice, non-periodic functions often
arise. This is particularly so in protection, where transient waveforms have to
be processed. Therefore the question arises as to how non-periodic functions
can be treated.
The first step in answering this question is to note that a non-periodic
function can be regarded as one having a periodic time interval T equal to
infinity (J1—» o°). It follows that/(/) = lim r _ 0 0 ^(/), where fp(t) is a replica of the
single function that is repeated after an assumed infinite time. If we assume that
the Fourier coefficients Fn of the function^(^) are known, then fp(t) itself can be
found from substituting eqn. 2.67 into eqn. 2.66. Therefore by assuming
tx = —772, we obtain

I fm
J -Til

or
1 j ; ([Tn.
(2J0)
ZJl
„=_« \J -r/2 F
Now, on allowing T—* °° , the following observations apply:

(i) the fundamental frequency a)0 = — = da)-^0 and na)o = a)

(ii) an infinite number of harmonics spaced by da)-—>0 are required to represent
the function / ( / ) . In essence, the time function is represented by a
continuous spectrum of frequencies all separated by an infinitesimally
small frequency space da).
It follows from the foregoing and eqn. 2.70 that

f(l)=\imfp(t)

Cm y«<v2jr
2(
H= —oo \J —772
>) T (2J1)

or

=
h \" eim da)
" (2-72)
 Mathematical background to protection algorithms 29

The term between the brackets in eqn. 2.72 is seen to be a function of frequency
and will be denoted by F((o), i.e.

F(o))=\ f(t)e-ja)ldt (2.73)

J — 00

The frequency function F((t)) is called the Fourier transform off(t). In terms of
the Fourier transform, the function/(/) can thus be written as

jQ>t
da) (2.74)

The time domain function f(t) and the corresponding frequency domain
function F(co) are commonly known as transform pairs, and any non-periodic
function of time can be transformed to a continuous frequency spectrum F(a))
via the Fourier transform integral of eqn. 2.73. The inverse Fourier transform
likewise inverts the process by transforming a continuous frequency spectrum
into the corresponding time variation.

Example
Find the Fourier transform of the rectangular pulse shown in Figure 2.5(a).

Solution
The pulse shown in Figure 2.5(a) can be expressed mathematically as

=[ lo
elsewhere
l
The Fourier transform of PT(t) is obtained from eqn. 2.73 as

PT{w)=(" Pt(l)-j°"dt
J -00

fr/2 I" C-J"» Tr/2

e-Ja" dt=\
J-r/2 L-^J-r/2

which can be converted to the form

sin(o>r/2) (DT
Pr{<»)=x a)x/2 =rsincy (2.75)

Figure 2.5(^) shows how PT((o) varies with angular frequency. It is interesting to
note that the shorter the pulse, the more quickly the Fourier transform falls to
30 Mathematical background to protection algorithms

zero, i.e. as the pulse duration x is reduced, the magnitudes of spectral

components also reduce and the distances between consecutive crossings of the
ft>-axis become larger. This is as expected since a reduction in the pulse width is
accompanied by a reduction in signal energy.

2-7 Walsh function analysis [12, 13, 14]

The Fourier series approach enables a periodic function to be represented by a

summation of discrete sinusoidal components. An alternative approach enables
a periodic function to be represented by a series of orthogonal square waves
known as Walsh functions. The Walsh function is consequently useful where
signals take the form of logic pulses, which are fundamentally rectangular in

Pr(t)

-1/2 x/2

Figure 2.5 (a) Non-periodic (transient) pulse and (b) corresponding Fourier transform
 Mathematical background to protection algorithms 31

nature. Conversely, it is relatively easy to generate square (or rectangular)

pulse trains, and it is consequently relatively easy to construct or analyse a
given waveform using the Walsh technique.

2.7.1 Definition of Walsh functions

The Walsh function of order k (written as Wal(&, t/T)) is a time function having
a magnitude of either 4-1 or — 1 that crosses the time axis k times per unit
interval T (i.e. the number of times it changes its sign per unit interval T is
equal to k).
Figure 2.6 shows the first eight Walsh functions of integral index k for
0 ^ ^ T. The number of sign changes per unit time is called the sequency

S=k/T

2.7.2 Some fundamental properties of Walsh functions

(a) Index and timescale
Doubling the index of a Walsh function is equivalent to compressing the
timescale by a factor of half, or

Wal(2*, t/T) = Wal(*, 2t/T) (2.76)

(b) Multiplication
When two Walsh functions are multiplied together, another Walsh function is
produced such that

Wal(A, t/T) • Wal(*, t/T) = Wal(A<g>A;, /) (2.77)

where h®k denotes a number whose binary numeral has 0s in those positions
where the binary numerals for h and k are alike and has Is where they are
different. For example

Wal(3, t/T) Wal(5, //7) = Wal(6, t/T)

Because 3 (in decimal) =011 (in binary), and 5 (in decimal) = 101 (in binary),
3 ® 5 = 110 (binary) = 6 (decimal).

(c) Symmetry
With reference to Figure 2.6, it can be concluded that

/ T-t\
Wai *, - = - = (-1)* Wai(t, t/T) (2.78)

Thus Wai (A;, t/T) has even or odd symmetry about /=7V2 depending on
whether k is even or odd.
32 Mathematical background to protection algorithms

1.

Wal(O, t/T)

-1

Wa!(1,t/T)

Wal(2, t/T)\1 1
T

Wal(3, t/T)\) I
T

-1

Wa!(4, t/T)

-1

Wal(5, t/T)

-1
1

Wal(6, t/T)

-1

Wa!(7, t/T)

-1

Figure 2.6 The first eight Walsh functions of integral index kfor
 Mathematical background to protection algorithms 33

2.7.3 Discrete representation of Walsh functions

In digital applications, it is useful to represent Walsh functions in discrete form.
Let Nbe the number of samples per unit interval, i.e., the length of the discrete
Walsh function is N. If JV= 2P, then the discrete Walsh function Wal(&, j/T) can
be defined as

with
jr = 0, 1,2, ,N-\

* = 0, 1,2, , JV—1

where II is the factorisation symbol in which the parameters that follow are
multiplied together. kn j n are binary bits of k, j or

k = ^kr2r and j = ^jr.2r (2.80)

r=0 r=0

This definition can be best illustrated by the following example.

Example
Find the fourth element of the Walsh function having the order of 5 and length
JV=8, i.e., Wal(5, 4).
Solution

therefore
P=3

k = 5 (in decimal)

= 101 (in binary) = 1 X22 + 0x2* + l X 2°

*o=l,*, = 0, * 2 = 1
Similarly
j = 4 (in decimal)
= 100 (in binary)

hence
34 Mathematical background to protection algorithms

Now Wai(5, 4) can be found from eqn. 2.79

3

Wal(5, 4 ) = J | (
r=0

= (-l)°-(-l)°.(-l)1 (because Jo = ^ = 0)

= -1

Using this procedure it is possible to construct an 8 X 8 matrix for an 8-length

eight function:

j 0 1
*
1 1 1 1 1 0
1 1 -1 -1 -1 1
1 1 -1 1 2
Wal(kJ) = 1 1 1 -1 3
1 -1 -1 -1 4
1 -1 1 1 5
1 -1 -1 1 -1 6
1 -1 -1 1 7

2.7.4 The Walsh series

In terms of Walsh functions, a periodic function say/(/) can be expanded over

the interval (tu tx + 7) as follows:

,tlT)
(2.81)

where the Walsh coefficient Wk is given by

•4/: /(/) Wai (A:, t/T) dt

(2.82)

where Wk is the A:th coefficient of the Walsh series and Wai (A:, t/T) is the A:th
order Walsh function. This technique mirrors the previously described method
of constructing a periodic waveform by a number of sinusoidal components
using the Fourier series technique.
 Mathematical background to protection algorithms 35

2.8 Relationship between Fourier and Walsh coefficients

In this Section we establish the interrelationship between Fourier and Walsh

coefficients, i.e. the transformation matrix which transforms a Walsh coefficient
vector W into its equivalent Fourier coefficient vector F.
Let us first write a Fourier series defined by eqn. 2.59 in terms of the Fourier
coefficients Fj(i = 0, 1, 2, . . .), such that
/(/) = Fo + V 2 F, sin a)ot + V 2 F2 cos a)ot + V 2 F3 sin 2a)ot + V 2 F4
cos 2a)ot+ (2.83)
It is easily verified that

/(/) sin na)ot dt, n= 1, 2, . . . (2.84)

^_
an = V2
V2 [r
F2n = vT7T = -Y /(/) cos nct)ot dt, n= 1, 2, . . .
^ Jo
If the same function f(t) is expanded using Walsh series, we have:

(2.85)

Now it is possible to calculate Walsh coefficients {Wk) by substituting eqn. 2.83

into eqn. 2.82 such that

l
W0 = - \ [Fo + V 2 F, sin (o0t + V 2 F 2 cos a)ot + V 2 F 3 sin 2(o0t

. . .]Wal(0, tlT)dt

x=
~f\ J €O S

l, tlT)dt

i r/i+r
Wk = - I [Fo + V 2 F, sin a)ot + V 2 F 2 cos coot + V 2 F 3 sin 2a)ot
36 Mathematical background to protection algorithms

Using matrix notation, these equations can be written as

A oo A)l ^02 . . AOh Fo

wt -410 Al2 . •• A\h
w2 ^20 A22 A2k (2.86)

wk Fk

or
W=AF
where
1 f'. + r

V 2 f'i + r
A:, 2m-1 =—— I sil Wal(*, t/T)dt (2.87)

V2 pi
cos(mo)0)t ,t/T)dt,m=l,2,3,...
J '1
Eqn. 2.87 can be simplified by taking the lower time limit / | = 0 , and
normalising the time with respect to the period T, such that

tf=-,dt=Tdtf (2.88)

When eqn. 2.88 is substituted into eqn. 2.87 (replacing the subscript 2m by K
and simplifying) we obtain:

Au\ — Wal(A:, t')dt'

I An(Kn)t' Wal(*, / ' ) * ' (2.89)

J0

I ^r)/'Wal(*,/')
J
By direct substitution of the Walsh functions Wai (A:, / ' ) , shown in Figure 2.6, it
is possible to calculate the elements of the matrix A having any given size. For
example the elements of a 12 X 16 A-matrix have been evaluated using this
approach to obtain the matrix eqn. 2.90 (see page 37).
As indicated in eqn. 2.86, the relationship between Fourier and Walsh
coefficients is given in matrix form as
W=AF (2.91)
 Mathematical background to protection algorithms 37

CD
CM

2 O O O O O O O O O O O O O
.-* O O O O O O O O O O O O O
CD CO
CM
p °1
o d d
o o I o o o I o o o I o o
CO
CD r~5 00
CM • ^O
^ O CM
odooo I ooodooo

CM
-* o o o o
1I o o o o o o o o
^

^ O O O O O O O O O O O O O

o ^ o
22 © S
~ o o o o o oCO I o oO
too o o o
<£5
CD O O O O O OJ O O O O O O O
o
O O O O O O O O O O O O O
o
O O O O O O O O O O O O O

^ S CO
O r^. ^
CD O O I O O O O O O O O O O
O CM
CO t-"^ O
m oooooo* o o o I o o o
O CO
CD CO
Th O O O O O O O O O O O O O
CO

CO O O O O O O O O O O O ! O

O r^ H
CD CO O
o o o o o o o o o o I oo
CO ^
^. t^- t^»
§ <« p
CD d d
o d o o o Io o o I o o o
— o o o o o o o o o o o o
38 Mathematical background to protection algorithms

where W is the Walsh coefficient vector, F is the Fourier coefficient vector, A is

the matrix that transforms F-vectors into W-vectors and whose elements are
defined by eqn. 2.90.
Fourier coefficients can likewise be determined from Walsh coefficients by
using the fact that A~l =AT. Thus the inverse relationship is
F=ATW (2.92)
T
where A is the transpose of matrix A.

2.9 References
1 KREYS^IG, E.: 'Advanced engineering mathematics' (5th Edition, John Wiley &
Sons, 1983)
2 WYLIE, C.R.: 'Advanced engineering mathematics' (4th Edition, McGraw-Hill,
1975)
3 LaFARA, R.L.: 'Computer methods for science and engineering' (Intertext Books,
1973)
4 HILDEBRAND, F.B.: 'Introduction to numerical analysis' (2nd Edition,
McGraw-Hill, 1974)
5 GERALD, C.E.: 'Applied numerical analysis' (4th Edition, Addison-Wesley, 1989)
6 APOSTOL, T.M.: 'Calculus—introduction with vectors and analytic geometry',
Vol 1 (3rd Edition, Blaisdell Publishing Company, 1962)
7 RALSTON, A.: 'A first course in numerical analysis' (2nd Edition, McGraw-Hill,
1978)
8 KAPLAN, W. 'Advanced mathematics for engineers' (Addison-Wesley, 1981)
9 STANG, G.: 'Linear algebra and its applications' (Academic Press, New York, 1976)
10 MAYHAN, R.J.: 'Discrete-time and continuous-time linear systems' (Addison-
Wesley, 1984)
11 McGILLEM, C D . , and COOPER, G.R.: 'Continuous and discrete signal and
system analysis' (2nd Edition, Holt, Rinehart & Winston, 1984)
12 BLACKMAN, N.M.: 'Sinusoids versus Walsh functions' Proc. IEEE, 1974, 62,
pp. 346-354
13 SCHREIBER, H.H.: 'Bandwidth requirements for Walsh functions' IEEE Trans.,
1970, IT-16, pp. 491-493
14 KENNET, B.L.N.: 'Note on the finite Walsh transform' IEEE Trans. 1970, IT-16,
pp. 489-491
 Chapter 3
Basic elements of digital protection

3.1 Introduction
Operating voltages and currents flowing through a power system are usually at
kilovolt and kiloampere levels. However, for digital processing, it is necessary to
reduce the primary measurands to manageable levels. Therefore, the analogue
signals are converted to digital form, thereby allowing subsequent digital
processing to be performed to determine the circuit state.
In this Chapter the basic principles underlying the conversion of analogue
signals into equivalent digital forms will be explained. We shall also explain the
essentially common features of various digital relaying schemes, other detailed
aspects being discussed in later chapters.

3.2 Basic components of a digital relay

Any digital relay can be thought of as comprising three fundamental sub-
systems (Figure 3.1):
(i) a signal conditioning subsystem
(ii) a conversion subsystem
(iii) a digital processing relay subsystem.
The first two subsystems are generally common to all digital protective
schemes, while the third varies according to the application of a particular
scheme. Each of the three subsystems is built up of a number of components
and circuits, as discussed in detail in the following sections.

3.3 Signal conditioning subsystem [1-5]

3.3.1 Transducers
Primary power system currents and voltages are usually relatively high. Before
it is possible to bring these signals to protective relays, they must therefore be
reduced to much lower levels. Conventionally, currents are reduced either to
5 A or 1 A and voltages are reduced to 110 V or 120 V. This is normally
achieved by using primary current and voltage transducers (CTs and VTs). In
digital relays, however, current and voltage magnitudes are both further
40 Basic elements of digital protection

transducer

surge
protective
circuit

signal
conditioning
LP
subsystem
filter

analogue
multiplexer

sample
hold
circuit

A/D
conversion converter
subsystem

digital
multiplexer

digital
processing D/O D/l memory CPU
relay
subsystem
D/A

trip remote location

signals (s) data

LP = low pass
A/D = analogue-to-digital
D/A = digital-to-analogue
CPU = central processor unit
D/l = data input
D/O = data output
Figure 3.1 Basic components of a digital relay
 Basic elements of digital protection 41

reduced using auxiliary transducers and/or mimic impedances within the

relays to suit the requirements of the components used.
Ideally the current transformer would reproduce a perfectly scaled down
version of the primary signal on its secondary side. Practical transformers
reproduce the secondary current with some error, because these devices
incorporate 'non-ideal' elements. The worst condition occurs when the ironcore
saturates during faults. The degree of signal distortion and the time after a fault
at which it occurs is heavily dependent on the total burden connected to the
primary current transducers. However, in most modern practical applications,
the overall burden is such that the current signal distortion is small during the
measuring period. In cases where this is not so, it is desirable to establish the
effect on performance and, if necessary, incorporate means within the relay
software to ensure that integrity of measurement is maintained. Some work on
compensating for current transducer saturation is to be found in the literature,
although, in most applications, this is unnecessary. In what follows, the effect of
current transducer saturation will, for brevity, be assumed negligible.
Electromagnetic VTs generally produce a very high-fidelity voltage signal,
though in fact these are rarely used at system voltages above typically 100 kV.
At higher voltages, the use of capacitor voltage transformers (CVTs) is
commonplace. Unfortunately, the transient response of CVTs varies widely
according to the type of transducer involved and the nature of the total
connected burden. In high-speed relaying applications in particular, account
needs to be taken of the rather poor fidelity of the voltage signals that emanate
from CVTs, there being many examples in the literature by which the digital
algorithm compensates for such effects. A detailed consideration of such
techniques is not necessary in a text of this type.
3.3.2 Surge protection circuits
The current and voltage from the secondaries of the CTs and VTs is connected
to surge protective circuits, which typically consist of capacitors and isolating
transformers (Figure 3.2). Zener diodes are also commonly used to protect
electronic circuits against surges, though their placement depends on the exact
physical circuit arrangement used. In practice, it is common to convert the
secondary current measurands into low-level voltage signals by means of a
suitably connected burden and/or current-to-voltage amplifier arrangement.
The latter normally use careful screening techniques and are often accommo-
dated inside separately screened self-contained modules separated from the
digital signal processing hardware.
3.3.3 Analogue filtering
It is normally necessary to perform analogue filtering of the signals received
from the CTs and VTs. In practice, the amount of filtering depends on the data
requirements of a particular digital relay. Such filtering is usually performed
using low-pass filters to remove unwanted high frequencies before sampling. In
fact, as will be shown in Section 3.4, the sampling theorem requires that
42 Basic elements of digital protection

isolating
surge capacitor transformer

to LP filters

Figure 3.2 Simple surge protective circuit

analogue signal components above a certain frequency (which in turn is related

to the digital sampling frequency) be attenuated to avoid errors in subsequent
digital processing. It is this 'anti-aliasing' function that is importantly fulfilled
by the analogue low-pass filters, which must be designed with a cutoff frequency
(fc) that performs satisfactory signal component rejection above a given
frequency. Figure 3.3(a) shows the characteristics of an ideal low-pass filter,
which transfers signal components of frequencies below the cutoff frequency
with zero attenuation, while components above the cutoff frequency are
attenuated to zero. The effect of introducing a practical low-pass filter is shown
in Figure 3.3(b), from which it can be seen that in practice it is not possible to
achieve such a pronounced transition between the pass and stop bands.
The dynamic characteristic of the low-pass filters, as well as their steady-state
characteristics, are important. Among the more important factors are
(i) the rise time, which gives an indication of how long it takes the output of a
low-pass filter to traverse its final value following a step input
(ii) the overshoot, which indicates by how much the filter output will exceed
its steady-state value on the initial response to a unit step input
(iii) the settling time, which is an indication of how long it takes a given filter to
settle at its steady-state output value.
 Basic elements of digital protection 43

output voltage
input voltage
A
1

Figure 3.3 Characteristic of low-pass filters; (a) ideal filter response, (b) practical filter
response

All the above features play a part in the overall dynamic response of digital
relay systems. In particular, in systems where a very high speed decision is
required, it is particularly important to ensure that the low-pass filter is
designed to have a cutoff frequency that gives an overall performance that is not
degraded by long filter delays [6].

3.3.4 Analogue multiplexers [7]

In digital relaying applications, it is usually necessary to use an analogue
multiplexer. The concept of multiplexing has its origins in communications
engineering. An analogue multiplexer is a device that selects a signal from one
of a number of input channels and transfers it to its output channel, thereby
permitting the transmission of several signals in a serial manner over a single
communication channel. The principles of multiplexing are thus as shown in
Figure 3.4, in which the solid-state multiplexer is likened to a multi-terminal
rotary switch.
44 Basic elements of digital protection

inputs 4
output
5

6
7
Figure 3.4 Principle of multiplexing

3.4 Conversion subsystem

3.4.1 The sampling theorem [8]
The sampling theorem states that a band-limited signal can be uniquely
specified by its sampled values if and only if the sampling frequency is at least
twice the maximum frequency component contained within the original signal,
i.e.
f^Vm (3.1)
where/ is the sampling frequency andX is the maximum significant frequency
within the signal sampled.
The frequency component at half the sampling frequency is known as the
Nyquist frequency fN, or

/*=I/2 (3.2)

The sampling process is effectively achieved by connecting an analogue signal

f(t) to the data acquisition system by means of a fast acting switch, which closes
for a very short time but remains open for the rest of the period (Figure 3.5(a)).
This operation can be modelled by a multiplier (Figure 3.5(b)), where/(/) is the
band-limited analogue signal to be sampled, and s(t) is known as a sampling
function. The sampling function is, therefore, made of a train of pulses
alternating between a value of + 1 and 0. It is thus defined as follows:

W-*T.) (3.3)

The output of the multiplier f(t) is then

f(nTs)d(t-nTs) (3.4)
 Basic elements of digital protection 45

f(t) «s(t)

-3T S -2T S -T S 0
• t 11 t,
T s 2TS 3T s 4T s t
M CDs
M,2C0s 03

Fs((0) practical low-pass

fs(t)> filter
ideal filter

-3T S -2T S -T S 0
T
Ts 2TS 3TS
A"-
t \ 0
A.
Figure 3.5 Sampling processes; (a) the sampler, (b) representation of the sampler, (c)
functions f(t), s(t) andf(t), (d) Fourier transforms off(t), s(t) andfs(t)

Figure 3.5(c) shows the three time functions/(/), s(t) and f(t). It can be seen
that the sampling function f(t) consists of a train of pulses spaced equally by a
period equal to the sampling interval Ts. The amplitude of each sample is the
same as that of the original signal at the respective sample time nTs. Therefore,
the resulting samples are included within the original signal envelope, as shown
in Figure 3.5(c).
To show how and under what conditions/(/) can be uniquely reconstructed
from^(^), let us examine the spectra of the three time functions f(t), s(t), and
f(t). This is shown in Figure 3.5(</), in which it is assumed that the function
F(a)) contains no frequency above O)m. The spectrum of the sampling function is
thus given by

(3.5)
46 Basic elements of digital protection

where Q)s is the angular sampling frequency, given by (Os=2jzfs =

It will be noted that, in accordance with the convolution theorem, multiplica-
tion of two functions in the time domain is equivalent to their convolution in the
frequency domain. Therefore,

Fs(a>)=—F(to)*S(a>) (3.6)

The convolution of F(CD) with each pulse of S((o) produces Fs((o) displaced in
frequency steps equal to na)s. Thus

(3.7)

As illustrated in Figure 3.5(rf), it is important to note that there will be no

overlap of adjacent parts of Fs(o>) as long as

(3.8)

If this condition is fulfilled, the passing of the signal fs(t) through a low-pass
filter (see Figure 3.5(d)) with a bandwidth such that its cutoff frequency coc is
given by eqn. 3.9, results in a perfect digital reconstruction of the analogue
signal/(/).

(3.9)

3.4.2 Signal aliasing error [9]

If the sampling rate is chosen so that the sampling frequency is less than twice
the maximum significant frequency contained in the original signal, i.e. eqn. 3.1
is not satisfied, then it can be seen from Figure 3.5(d) that there will be an
overlap between adjacent parts of Fs((o). This leads to what is commonly known
as an 'aliasing error', which in turn causes an error in the analysis as a result of
the difficulty in distinguishing between low- and high-frequency components.
In other words, if the sampling rate does not satisfy eqn. 3.1, a low-frequency
component that does not actually exist in the original signal, would nevertheless
be apparent within the sampled signal. Figure 3.6 gives an illustration of the
aliasing phenomenon, in which the dotted line represents a low frequency that
does not actually exist in the original signal.
 Basic elements of digital protection 47

alias frequency signal signal sampled

i /
\ 1 \\
II ' \ i M 1
V l
I
U/

l
i \\
1 \
I 1 1 1
\ \ \\ 1
^ /

Figure 3.6 Simple illustration of aliasing phenomenon

3.4.3 Sample and hold circuit [10-12]

If the sampling switch of Figure 3.5(a) is replaced by a switch-capacitor
combination, each sample can be stored or held until the next sample is taken.
This process allows sufficient time to elapse for the subsequent process of
analogue to-digital conversion to be completed. Figure 3.7 shows the principles
of a simple sample-and-hold circuit applied to the signal derived from the
output of the signal-conditioning subsystem (see Figure 3.1).
Figure 3.7(a) shows a basic sample-and-hold circuit. The input is the
analogue signal/(/), which is sampled at a rate of l/Ts. Sampling is controlled
by the control waveform Vc, which closes and opens the switch. During the
closing time Tc, the capacitor is charged to the value/(/), while during the hold
time TH=TS— Tn the capacitor holds the sample value. The process of sample
and hold is best understood by referring to Figures 3.7(b), (c) and (d). There are
a variety of sample-and-hold circuits available in Litegrated-circuit (IC) form.
However, virtually all commercially available devices work on the principle
outlined, their performance being defined largely in terms of accuracy and
sampling rates.
48 Basic elements of digital protection

Vc

Figure 3.7 Sample and hold circuit principle (a) switching circuit, (b) analogue input, (c)
control waveform, (d) circuit output
 Basic elements of digital protection 49

3.4.4 Digital multiplexing [13-15]

A digital multiplexer is a network with a number of input ports channelled to a
single output. The input to these ports consists of digital words of information of
one or more bits.
Figure 3.8 shows a general representation of a digital multiplexer that allows
one of its three inputs to pass to the output side, such that
7=0 if [A,B] = [0,0]
Y=I\ \i[A,B] = [0,1]
(3.10)
Y=I2 X[A,B] = [\,0]
Y= 73 i{[A, B] = [l, 1]
where
Y=[Y0, Yu. . ,Yn] is the output
/I = [7 0lJ 7,,, . . . Ynl] is the first input
/2 = [702, ^i2> • • • ^2] is * n e second input
73 = [703, 7,3, . . . Yn3] is the third input
Figure 3.9 shows a typica]_Jogic diagram for a two-input-to-one-output
multiplexer. The input word 71, which consists in this case of four bits (Y0{, 7 n ,
y2i, 7 3l ) is transferred to the output 7 if the control signal (address) A is low.
On the other hand, when the address signal A is high, then the input 72 is
transferred to the output 7.

input 1

• output

input 3

A B

Figure 3.8 Digital multiplexer arrangement

50 Basic elements of digital protection

address input (A)

Yi (input 1)<

> Y (output)

Y 2 (input 2){

Figure 3.9 Logic circuit for two input/one output multiplexer

3.4.5 Digital-to-analogue conversion [12-16]

Consider first the method of converting a digital representation to analogue
form. The reason for this is that D/A converters are frequently used in
analogue-to-digital converters. The principles involved are well illustrated by
considering the circuit of Figure 3.10. This may be used to convert a four-bit
binary parallel digital word W3W2WlW0, where any Wt is either 0 or 1, to an
analogue voltage that is proportional to the binary number represented by the
digital word. The logic voltages that represent any individual bit W{ are, as a
matter of fact, not connected to the converter but rather are used to control the
switches SOi Su S2 and $3. Therefore if W^= 1, then S{ is connected to VR while if
Wj = 0> then S{ is connected to ground.
The values of the circuit elements are chosen so that successive resistors on
the input side are related in powers of two, and the individual resistors are
inversely proportional to the numerical significance of the appropriate binary
digit (Figure 3.10).
 Basic elements of digital protection 51

VR
O

least significant bit (LSB)

i
R/2

1 O

R/2 2

s2
1
R/2 3
most significant bit (MSB)

1
Figure 3.10 Basis of simple digital-to-analogue converter arrangement

Remembering that the input impedance of the operational amplifier is very

high, the currents Is and If are roughly equal to each other. The relationship
between the output voltage of the operational amplifier (Fo) and the digital
input can be found as follows:

Is=(23W3 + 22W2 + 2lWl + 2°W0)VR/R (3.11)

R (3.12)
It is clear from eqn. 3.12 that the output signal voltage Vo is directly
proportional to the binary number W3W2WX Wo- For illustrative purposes assume
that the binary input array corresponds to 15 (decimal), i.e. W3W2WlW0= 1111;
in this case the output is proportional to (23 + 22 + 2* + 2°) = 15 as required.
Similarly, an input word of 0011 would give an output proportional to
(0 + 0 + 2 1 + 2°), corresponding to the input number 3.
52 Basic elements of digital protection

3.4.6 Analogue-to-digital conversion

The block schematic of an analogue-to-digital converter is shown in Figure
3.11, where Va is the analogue voltage input and W \s the digital output array.
For simplicity, a four-bit word converter giving an output array WZW2WXWQ is
shown. In the field of digital integrated electronics, many methods have been
developed for converting analogue signals to digital form, the most commonly
used being counter-controlled converters, dual-slope converters and parallel-
comparator converters.

digital output

O W3(23)
u
analog input A/D r\
converter
^\
U
Va
r\ W0(20)
•O
U

Figure 3.11 Block diagram of analogue-to-digital converter

3.4.6.1 Counter-controlled converter

This technique represents one of the simplest methods of A/D conversion.
Figure 3.12 shows the basic arrangement involved. It consists of three main
components: a counter, a D/A converter and an analogue comparator. Its
operating principles may be summarised as follows. The counter is set to zero at
the beginning of each cycle, which causes the output of the D/A converter Vc to
be zero. This latter voltage is compared with the input voltage Vt fed by the
sample-and-hold circuit, using the comparator whose output will be either 1 or
0 depending on the relative magnitudes of Vt and Vc. If Vt> Vc then the
comparator output will be 1, which is used to enable the AND gate. This in turn
allows the clock pulses to enter the counter.
Each pulse entered in the counter causes the output voltage of the A/D
converter to increase by a single step of 1 V, as shown in Figure 3.12(^). As soon
as Vc becomes greater then Vh i.e. V(< Vn the output of the comparator will be
zero. This disables the AND gate, and the clock pulses are thereafter prevented
from reaching the counter, which then stops. The output of the A/D converter
is then read from the output terminals of the counter. The example illustrated in
Figure 3.12 involves an analogue input voltage that is marginally above 5
quantum levels, and is identified as the equivalent binary number 0101. The
resolution of the counter is controlled by the gain on the D/A converter, while
the dynamic range is largely dictated by the number of binary stages within the
counter. In relay applications quantum levels between 5 juV and 300 juV are
fairly common, the precise level being a function of dynamic range require-
ments. This type of counter is useful in some applications, but has limited speed
for a given resolution.
 Basic elements of digital protection 53

4-bit counter

8 4 2 1 digital
output

comparator,
output = 1 whenVj>Vc

D/A converter

analog
input

i
— i
i
4
I
i
-
counter stops at 0101

3 -h Vc

2 -

1 -
i—i—i—i—>
2 3 4 5 6
multiples of clock time step

Figure 3.12 Counter-controlled converter (a) circuit arrangement (b) illustrative response

3.4.6.2 Dual-slope converter

This converter basically consists of an analogue integrator, a comparator and a
counter. Its operating principle is best understood by reference to Figure 3.13.
54 Basic elements of digital protection

analogue-integrator

Vo<0
0 Vo>0

W0W1W2W3

digital output

Figure 3.13 Dual-slope converter

At time £ = 0, the switch Sx is connected to A, and the sample hold voltage Va

is applied to the analogue integrator, which integrates Va for a fixed time Tx.
After that the switch Sx is thrown to B, which in turn connects a negative
reference voltage — Vr to the integrator. Therefore the reference voltage will be
integrated and the output of the integrator starts to move in the positive
direction (Figure 3.13(£)). As soon as the integrator output voltage Vc reaches
zero, the gate G is disabled, and this in turn stops the clock pulses from reaching
the counter, thereby stopping the count. The number of clock pulses admitted
to the counter is thus proportional to the magnitude of the input voltage Va
which is thereby converted to a digital representation. In practice, the reset
 Basic elements of digital protection 55

voltage Vr is arranged to stop the count quickly. This type of converter is

characterised by a high accuracy, but it does have a relatively slow conversion
rate, which limits its application in some digital relaying devices.

3.4.6.3 Parallel-comparator converter

Figure 3.14 illustrates a three bit parallel-comparator A/D converter. It
consists of comparators C, to C7, a register of seven flipflops and a decoder.

comparators registers

Figure 3.14 Parallel-comparator converter

56 Basic elements of digital protection

The input ranges from 0 to Fo and is divided into eight reference segments, six
of which have a value of VJ1 the two end segments having a value of Fo/14
each. If an input voltage Vh with magnitude ranging from 0 to Fo/14, is applied
to the converter, then the outputs of all comparators Ci-C 7 are set to logical 0.
The digital representation of this at the output of the converter is seen to be 000.
This is equivalent, as required, to a zero analogue voltage. It will be noted that
an error of Fo/14 is introduced on account of the input being equal to one level
of resolution (F o /14). This is often called a 'quantisation' error.
If the input voltage magnitude is between Fo/14 and 3FO/14, then the output
of all the comparators is zero, except the first one which will be 1, i.e.
CiC2C3C4C5C6C7= 1000000. The output of the comparators is transferred to the
register flipflops at the occurrence of a clock pulse and the register output is
converted by the decoder to a three-bit unipolar binary code.

3.5 Digital relay subsystem [5, 6? 17]

The digital relay subsystem comprises both hardware and software. The
hardware largely consists of a central processor unit (CPU), memory, data
input and output (I/O). The software is influenced by two major factors. The

start/

Dl - - digital data input

elimination of low-frequency
" " and higher harmonic
components

" " operating principles

determination of trip zone

" " output - trip instruction

Figure 3.15 Flow chart for the software of a digital protective relay
 Basic elements of digital protection 57

first of these is the operating principles and performance required, which either
leads to the development of a special algorithm or the implementation of an
existing one. This factor greatly affects the determination of the sampling
frequency, type of hardware structure and the data input hardware system.
The second factor is the digital filtering. Subharmonics, as well as high
harmonic components, can cause false tripping, failure to trip and variation in
protective relay performance. The operating principle and digital filtering must
in general provide for a wide application range and requirements relating to
speed of response to system faults, the latter being influenced by the necessary
computation time.
Figure 3.15 shows an example of a flow-chart for the software of a typical
digital protective relay. The algorithms used and the software required vary
significantly according to the application, and much of the work of later
chapters will be concerned with specific algorithms for meeting a variety of
protection performance and application requirements.

3.6 The digital relay as a unit

There is an inevitable tendency for relay manufacturers to develop standardised

hardware, which can be used in conjunction with suitably developed software to
meet a variety of production requirements and applications. Figure 3.16 shows
a block diagram of a standardised digital relay unit, and Table 3.1 gives a
typical basic specification of such a unit.

Table 3.1 Specification of a typical standardised digital relay unit [5]

Sampling frequency 720 600 sample/s

Data acquisition Data length 12 bit

Data CPU Device bipolar

processing word length 16 bit
language assembler

Memory IC memory

Tap setting Nonvolatile IC memory (up to 256 words)

 auxiliary digital relay unit
transformer .—
unit

auxiliary
relay
unit
§•

^
2

fil = filter S/H = sample-and-hold circuit

MPX = multiplexer A/D = analogue-to-digital converter
CLK = clock IDB = input data buffer
CPU = central processor unit Ry = trip relay

Figure 3.16 Digital relay unit

 Basic elements of digital protection 59

3.7 References
1 WRIGHT, A.: 'Current transformers—their transient and steady state performance'
(Chapman & Hall, 1968)
2 STALWESKI, A., and WELLER, G. E.; 'Novel capacitor divider voltage sensors for
high-voltage transmission systems', Proc. IEE, 1979, 126, pp. 1186-1195
3 'Computer relaying', IEEE Tutorial course 79 EH01487-PWR, 1979
4 GILCHRIST, G.B., ROCKEFELLER, G.D., and UDREN, E.A.: 'High speed
distance relaying using a digital computer, part I—system description', IEEE Trans.
1972, PAS-91, pp. 1235-1243
5 SEKINE, Y., HATATA, M., and YOSHIDA, T.: 'Recent advances in digital
protection', Elect. Power & Energy SysL, 1984, 6, pp. 181-191.
6 JOHNS, A.T., and WALKER, E.P.: 'Genesis and evolution of advanced protection',
Power Eng.J. 1991, 5 (4), pp. 177-187
7 CAHILL, J.J.: 'Digital and microprocessor engineering' (Ellis Horwood, 1982)
8 MAYHAN, R.J.: 'Discrete-time and continuous-time linear systems' (Addison
Wesley, 1984)
9 NATARAJAN, N.A.T.: 'Discrete-time signals and systems' (Reston, 1983)
10 TAUB, H., and SCHILLING, D.: 'Digital integrated electronics' (McGraw-Hill,
1977)
11 STANIER, B.J.: 'Electronic and integrated circuits' (Adam Hilger, 1985)
12 SCHILLING, D., and BELVE, B.: 'Electronic circuits—discrete and integrated'
(McGraw-Hill, 1979)
13 ibid, 3rd edition (McGraw-Hill, 1989)
14 BOOTH, T.L.: 'Digital network and computer systems' (2nd edition), (John Wiley,
1978).
15 HILL, F.J.: 'Introduction to switching theory and logical design' (John Wiley, 1981)
16 MALVINO, A.P.: 'Digital computer electronics' (McGraw-Hill, 1977)
17 MAKINO, J., and MIKI, Y.: 'Study of operating principle and digital filters for
protective relays with digital computers', IEEE PES Winter Power Meeting, New
York, January 1975, 75CHO990-2 PWR, Paper No. C75 197-9, pp. 1-8.
 Chapter 4
Sinusoidal-wave-based algorithms

4.1 Introduction
The algorithms covered in this Chapter assume that the post-fault current and
voltage waveforms are sinusoidal. This assumption is not, of course, generally
valid, particularly when EHV or UHV network applications are involved.
However, in practice, the signals processed are often prefiltered and, in lower
voltage distribution systems in particular, the waveforms often very quickly
attain a nominally sinusoidal form. Historically, algorithms developed for use in
applications where the signals processed are nominally sinusoidal were the first
to emerge [ 1, 2]. Most of the early work involved applying the technique to the
calculation of transmission-line fault-loop impedances. However, the methods
are equally applicable to determining the magnitude and phase of relaying
currents for differential protection of lines and plant.
All sinusoidal-based algorithms are designed to predict either the peak or
squared peak of the compared waveforms. They may be loosely classified into
two broad groups: those that use sample and first derivative (or first and second
derivative), and those that use two or three samples to predict the peak or
squared peak values.

4.2 Sample and first-derivative method [1,2]

4.2.1 Basic formulation
When a waveform is assumed to be perfectly sinusoidal, its peak can be
predicted from any of its samples. Consider for example two signal measurands,
Si(t) and s2(t) say, which take the form of eqns. 4.1 and 4.2 respectively.
Ji(/)=F, sina>ot (4.1)
s2(t)=V2sin(a)Qt + 6) (4.2)
where Vx and F2 are the peak values of the signal waveforms. By taking the time
derivative of eqn. 4.1, we obtain
s\(t) = a)0Vl cos a)ot (4.3)
d
(where, for brevity, s[(t) is written for —
at
 Sinusoidal-wave-based algorithms 61

When eqns. 4.1 and 4.3 are combined together, we obtain eqn. 4.4, which
defines the peak values of relaying signal ${t.

(4.4)

Similar equations to eqns. 4.1-4.3 can be derived to describe the peak value of
the second relaying waveform s2{t), i.e.

Determination of measured impedance

In distance protection, it is necessary to measure the apparent impedance of the
fault loop using signal measurands proportional to the voltage and current
monitored via the transformers located at the line ends. In this case, the
relaying signals can be conveniently described by eqns. 4.6 and 4.7:
sl(t) = v=Vsin (O0t (4.6)
s2(t) = i = I sin(a)ot + 6) (4.7)
In what follows, we will assume for brevity and convenience that the methods
described are applied to distance protection, although, as mentioned previous-
ly, they may equally be applied to differential protection. The basic measurands
will therefore be assumed to take the form of eqns. 4.6 and 4.7.
The measured impedance can be calculated from the magnitude and relative
phase of the voltage and current measurands given in eqns. 4.6 and 4.7. The
magnitude of the impedance is simply calculated as |Z| = V/I which, with the
basic measurands defined in eqns. 4.6 and 4.7 is given by
\Z\ ={[v2+ (v'/a>0)>]/[? + (i'Aoo) 2 ]}" 2 (4.8)
The argument (or phase) of the measured impedance (0) is equal to the angular
difference between the voltage and current signal. Thus, the argument of the
measured impedance is given by eqn. 4.9:
e=ev-e, (4.9)
where from eqns. 4.6 and 4.7, 0v=(Oot and di = a)ot+d.
The values of these angles can also be determined from a waveform sample
and its derivative. This is done by re-writing eqns. 4.6 and 4.7 in terms of 0 F and
6h which in turn results in
v=Vsm6v (4.10)
v'=(D0Vcosdv (4.11)
Eqns. 4.10 and 4.11 thus give
0v=ten-l(va)0/v') (4.12)
Similarly, 0/ can be found as
0,= tan" 1 (*tt/0/O (4.13)
62 Sinusoidal-wave-based algorithms

Substituting eqns. 4.12 and 4.13 into 4.9, we thus obtain the argument of the
measured impedance in terms of the instantaneous values of the measurands v
and i, together with their derivatives:
x
-i2in-\ia)Jil) (4.14)

4.2.2 Calculation of an approximation to the signal derivatives

We have shown in the previous Section that the magnitude (|Z|) and angle (6)
of the impedance depends on sample values as well as the derivative of the
current and voltage measurands. Therefore, it is essential to know how to
determine the required derivatives from measured samples. This is achieved by
using the numerical techniques given in Chapter 2, which involves using a
series of expressions for forward, central or backward differences. However, for
real-time implementation, it is often advantageous to use backward differences
because existing samples can be used for immediate derivative calculation.
By substituting a particular voltage sample vk and the corresponding voltage
derivative v'k forfk a n d / / , respectively, in eqn. 2.27, we obtain

(v+V+..)vk (4.15)

where h = At is sampling time interval and

Using only the first term in eqn. 4.15 leads to

1
v'k^-^vk-vk.{) (4.16)

Using the first two terms likewise gives

{$vk*vk-x + vk.2) (4.17)

Similar equations can be used to estimate a current derivative (V). It will be

noted that, according to eqn. 4.16, it is possible to calculate the derivative
immediately after two samples become available, whereas the use of eqn. 4.17
requires that a third sample be available. The 'three sample' version does of
course give a generally more accurate estimate of the differential but requires
more computation.

4.2.3 Error analysis

To determine the error involved in numerically determining derivatives, let us
assume that the sinusoidal waveforms are sampled at discrete time intervals
h = At, with the actual sampling times being tk, tk+l, . . . . at which the
 Sinusoidal-wave-based algorithms 63

At/2 At/2

Figure 4.1 Calculation of voltage v0 halfway between samples taken at times tk and tk+I

corresponding, voltage samples are vh vk+\, Let t0 be a time halfway

between tk and tk+ly as shown in Figure 4.1. Then it follows that
1

and (4.18)
Av 1

Given that we are dealing with a sinusoidal waveform, the sampled value at
time tk can be written as vk = V sin (O0th which in turn is given by
^=Fsina>0(/0-A*/2) (4.19)
The voltage sampled one time interval later (vk+l) is likewise given by
vk+l=Vsin a)0(t0 + A//2) (4.20)
By substituting eqns. 4.19 and 4.20 into eqns. 4.18 we obtain
os(<Vy/2) (4.21)
sin(a>oA//2)]/A/ (4.22)
Because the value ofa)0At/2 is small, the sine and cosine terms can be readily
approximated by
co0At
cos - y - ~ 1 (4.23)
64 Sinusoidal-wave-based algorithms

a)0At a)0At
sin—~-y- (4.24)

By substituting eqns. 4.23 and 4.24 into 4.21 and 4.22, we obtain
vo^V sin(ttM0) [1 - (a)0At)2/8] (4.25)
and
vf0^a)0Vcos a)oto (4.26)
The estimated value of the peak of the voltage signal using sampled data at
t0 (Fo) can be found using eqn. 4.4:

When this equation is combined with eqns. 4.25 and 4.26, we obtain

or
F o « F[l - (colAt2 sin2 (o0t0)/S] (4.28)
It follows that the percentage error (e%) associated with the estimated value of
the peak of the voltage waveform will be

2
Eo/o =
a)i(At)
— sin2 m ^ x 10Q (4 29 ^
o

Eqn. 4.29 indicates that the error in the estimated peak value is a function of the
system frequency coOy the sampling interval At and the time at which the
estimation takes place. For a sampling interval At of 0.5 ms, the error is seen to
be a maximum when the estimation takes place at the extreme of the waveform,
i.e. at O)Qt0= ±nf, n= 1, 2, . . . On a 50 Hz system, this error is of the order of
0.3% (0.45% for 60 Hz systems).

4.2.4 Practical considerations

4.2.4.1 Decaying DC components
Practically, the current and voltage waveforms that immediately follow power
system faults are never pure sinusoids. This is particularly true for current
waveforms, which often contain a relatively large and slowly decaying DC
component. It follows that, in applying the above derived algorithm, care must
be taken in dealing with any DC offset components in the current signals, which
in practice can be transferred in full to the digital processing stage. This is
particularly so where the input circuit comprises an iron-cored transformer with
resistive burden.
 Sinusoidal-wave-based algorithms 65

The DC offset in the current signal can be reduced by using a mimic

impedance in the current transformer secondary, having an XIR ratio close to
that of the primary circuit. However, in practice it is not possible to match the
XIR ratio of the mimic to that of the power circuit for the complete range of
possible fault conditions, and some error is therefore caused by exponentially
decaying current and voltage components. Nevertheless, this technique can
greatly increase the range of applicability of techniques that assume purely
sinusoidal measurands.

4.2.4.2 Noise and digital smoothing

Another practical point of interest is that related to the error introduced by
noise in the input signals. Such errors can be reduced using digital smoothing of
sampled data. To avoid large delays caused by the process of digital smoothing,
three samples may be used with a first degree (straight line) approximation. To
meet these requirements, eqns. 2.56-2.58 can be used to calculate the values of
digitally smoothed samples. When this is done, the smoothed values of sampled
values corresponding to sampling instants k — 1, k and k 4- 1 are substituted into
the basic algorithm described. Denoting the smoothed values of (say) a voltage
by upper case letters, the values substituted would take the form of eqns. 4.30.

(4.30)

Eqns. 4.30 use forward, central and backward points, respectively. However, in
most applications of the algorithm it is preferable to use the backward
expression, because a smoothed value can be generated from the data already in
hand, thus speeding up the overall response.

4.3 First- and second-derivative method [3, 4]

Algorithms of this type generally reduce errors arising from subnormal
frequencies, as well as those due to slowly decaying DC transients. In essence,
they represent a refinement of the above detailed basic algorithms and play an
important part in applications where significant periodic and aperiodic compo-
nents are present in the signal waveforms.

43.1 Mathematical formulation

As previously, let the voltage and current waveforms be denoted by eqns. 4.6
and 4.7, i.e. v= Fsin (o0t and i = I sin(a>ot + 0).
Taking the first and second derivative with respect to time we obtain for the
voltage signal
v' = a)0V cos ct)ot (4.31)
2
v"=-a) 0V sin a>ot (4.32)
66 Sinusoidal-wave-based algorithms

Combining these two equations results in an equation for the square of the peak
of the assumed sinusoidal voltage:

[ ( Y l (4.33)
The corresponding equation for determining an approximation to the peak of
the current is likewise

/* = ! [(,-')*+(-Y1 (4.34)
The square of the magnitude of the measured impedance is

The argument of the measured impedance can be determined by using

procedures that are similar to those given in the previous section. Therefore,
from eqns. 4.31 and 4.32, the voltage angle is
l
6v=a)ot= tan

v
= — tan

and that of the current is

\a)oi'
It follows that the argument of the measured impedance angle (6) is given by

0 = tan" 1 '

First and second derivatives are commonly determined for use in this algorithm
by using divided differences (see Chapter 2). This is done by substituting the
voltage v and A* for the variables/and h, respectively, into eqns. 2.33 and 2.34,
in which case the following relationships are used:
1
( 4 3 6 )
**-i) *

1
^*+i-2» A + ^_i) (4.37)
where At is the sampling interval, and k + 1, k and k — 1 are subscripts referring
to a set of consecutive samples.
 Sinusoidal-wave-based algorithms 67

From eqns. 4.36 and 4.37, it can be seen that the main advantage of this
method is that a constant DC component has no effect since it is cancelled out
from all difference expressions. The effect of the non-zero frequency components
due to any exponentially decaying offsets is consequently minimised, and no
mimic burden or related technique is required in many applications. However,
the disadvantage associated with this algorithm is that higher-frequency
components, due in particular to fault-induced travelling waves, can produce
significant errors in the first- and second-difference equations, thereby causing
large errors in successive impedance estimates. A solution to this problem,
which is successful in many applications, is to use the smoothed form of sample
and first derivative version of the algorithm for processing the voltage signal
and the first and second derivative technique for the current. In this way,
maximum overall "noise" rejection can be obtained, thereby improving the
reliability of successive impedance estimates.

4.4 Two-sample technique [5]

In previous sections, the calculation of the magnitude and phase of the voltage
and current waveforms was achieved using derivatives. An alternative, which
avoids the need for finding differentials, involves manipulating two samples
taken at discrete instants of time from the signal waveforms.

4AA Prediction of values of peak (or magnitude) of signal waveforms

Let vh vk+l be voltage samples measured at times th tk+l respectively, and let At
be the sampling time interval. Then
vk=Vsin a)otk (4.38)
vk+1 = V sin 0)otk+, = V sin a)0(tk + At)
vk+l=Vsin a)otk cos (o0At + Vcos a)otk sin a)0At (4.39)
Substituting eqn. 4.38 into 4.39 and simplifying results in
vk+, — vk cos coQAt
: = V cos (o<fk (4.40)
v ;
sin Q)0At *
2 2
Adding the squares of eqns. 4.38 and 4.40, and noting that sin x + cos x = 1,
we obtain an equation for the square of the peak voltage:
v\ 4- v\+, — 2vkvk+1 cos Q)0At
(sin cooAt)2
The corresponding equation for the peak of the current is simply:

/2 _ 1i + ^1+
illI ~ 244+1
i l l : cos (OnAt
1!_ (A ±n\
(sin
68 Sinusoidal-wave-based algorithms

It can be seen that the values O)0At are fixed for any given nominal system
angular frequency and sampling interval At. They therefore appear as con-
stants in the evaluation of the peak values, which, in effect, simply weight the
sampled values used in the algorithm.

4.4.2 Determination of phase angle between waveforms

The phase angle between voltage and current waveforms can be calculated
using the above determined values of waveform magnitude (V and /) together
with the measured voltage and current samples.
Let 0 be the phase angle between the voltage and current waveform such that
the current sample ik is expressed as:
ik = I sin((DQtk + 0)
or
ik = / sin o)0tk cos 0 + / cos wotk sin 0 (4.43)
The current sample ik+{ (taken at a time tk-\- At) is likewise obtained in the form
^ + l = /[sin a)otk cos a)0At + co$ a)otk sin a)0At] cos 0
+ /[cos a)otk cos a)0At — sin a)otk sin a>0At] sin 0 (4.44)
From eqns. 4.38 and 4.40 we have
v
k
sin a)otk = ~- (4.45)

vk+l — vk cos a)0At

cos Q)otk = — (4.46)

Eqns. 4.43 to 4.46 can be manipulated to obtain a number of algorithmic

equations, the simplest of which is
~(*>*+i + W * ) cos o)0At)
rr7 , 2 1 \ (4.47

The above equation shows that the phase angle 0 can be determined from a pair
of voltage samples and current samples. Despite the apparent complexity of its
form, the above equation is in fact relatively easy to evaluate, given that, as part
of the overall determination of the measured impedance, the peak values of the
voltage and current measurands (V and I) must be evaluated anyway.

4.5 Three-sample technique [6]

This algorithm predicts the peak of the voltage and current measurands by
using three successive samples. It also calculates the resistive and reactive parts
of the impedance. This contrasts with the previously described algorithms,
 Sinusoidal-wave-based algorithms 69

which produce estimates of the magnitude and argument of the measured

impedance.
Let vh vk+u vk+2 be voltage samples measured at times tk, tk+l and ^+2> t n e n
vk= Vsin a)otk (4.48)
vk+l = V sin(a)otk +o)0At) (4.49)
v
»*+2 - sin(ay A + 2o)QAt) (4.50)
If ih ik+h ik+2 are the corresponding currents and if the phase angle between the
current and the voltage is 0, then
iA = / s i n ( a y * + 0 ) (4.51)
ik+l = I sin(a)otk + a)QAt + 0) (4.52)
14+2 = / sin(a>of* 4- 2cy0A/ + 6) (4.53)
By manipulating eqns, 4.48 to 4.53 and using appropriate trigonometric
identities, the following relationships can be proved:
iU i " 44+2 = / 2 sin2(a>0A/) (4.54)
v
~"~ k+2h+\ = — F/ sin 6 sin (W0A/ (4.55)
2
1 "Vk+2ik""Vkik+2 = 2 F/ cos 0 sin cw0A/ (4.56)
Now the resistive part (Rf) of the measured impedance is given by Rf=
(V/I) cos 6 so that, by using eqns. 4.54 and 4.56, we obtain

(4-57)

The reactive part (Xf) of the measured impedance is given by X/= (— F/7) sin 0,
and can likewise be readily determined from a three sample voltage and current
measurand set. This is done by means of eqns. 4.54 and 4.55, which leads to:

(4.58)

It should be noted that eqn. 4.57 shows that, as expected, Rf is independent of

sampling rate and therefore is not affected by the system frequency. On the
other hand, eqn. 4.58 shows that Xfi& dependent on the constant sin (OQAI; as
expected, the measured reactance is therefore proportional to the system
frequency.

4.6 An early relaying scheme [3, 4]

One practical distance protection scheme that uses the first- and second-
derivative algorithm was jointly developed by Westinghouse Electric
Corporation and Pennsylvania Power & Light Company. This is known as
Prodar 70, and it was one of the earliest developments which provided the
70 Sinusoidal-wave-based algorithms

AC
inputs A/D SPM P2000
signal A/Dsub inter-
, > conditioning system face <

k >

PSU

power >
supply
type- >
unit
writer
computer
sub-
PC system
programmers >
console

CO
contact
outputs
circuit- trip
breaker < relay
trip CKT

Figure 4.2 Schematic diagram for the basic components of the Prodar 70 protection system
(quoted from Reference 3)

major functions of distance protection as well as instantaneous over-current

protection and out-of-step blocking. Figure 4.2 shows a schematic diagram of
the scheme, in which voltage and currents derived from the CTs and VTs are
passed through a signal-conditioning unit and then to an analogue-to-digitai
subsystem where they are sampled and converted into digital form under the
control of a data buffer (known as scratch pad memory or SPM). The digitised
data are then transferred to the memory of the computer subsystem for
processing. If, as a result of the processing, a fault is detected, a signal is
initiated to trip the circuit breaker as well as for operating various event alarms.
Some peripherals are included in the system. Examples of these are a
KSR-typewriter and a programmed console to provide facilities for the logging
of data as well as a convenient means of execution control, software generation
and program checkout.

4.7 References
1 MANN, B J., and MORRISON, I.F.: 'Digital calculation of impedance for trans-
mission line protection', IEEE Trans. 1971, PAS-90, pp. 270-279
2 MANN, B.J., and MORRISON, I.F.: 'Relaying a three phase transmission line with a
digital computer', ibid, 1971, pp. 742-750
 Sinusoidal-wave-based algorithms 11
3 GILCHRIST, G.B., ROCKEFELLER, G.D., and UDREN, E.A.: 'High speed
distance relaying using a digital computer, part I—system description', IEEE Trans.
1972, PAS-91, pp. 1235-1243
4 ROCKEFELLER, G.D., and UDREN, E.A.: 'High-speed distance relaying using a
digital computer, part II—test results', ibid, 1972, PAS-91, pp. 1244-1252
5 MAKINO, J., and MIKI, Y.: 'Study of operating principles and digital filters for
protective relays with digital computers', IEEE PES Winter Power Meeting, New
York, January 1975, 75CHO990-2 PWR, Paper C75, 197-9, pp. 1-8
6 GILBERT, J.G., and SHOVLIN, R J.: 'High speed transmission line fault impedance
calculation using a dedicated computer', IEEE Trans. 1975, PAS-94, pp. 872-883
 Chapter 5
Fourier analysis and Walsh function
based techniques

5.1 Introduction
In this Chapter we present digital relaying algorithms which are based on
Fourier and Walsh analyses. From the point of view of this discussion, Fourier
analysis includes Fourier series and Fourier transform-based methods, while
Walsh analysis includes Walsh series only.
The basic assumption used in Fourier and Walsh series based methods is that
the waveform that results from a fault condition (voltage and/or current) is
assumed to be periodic within the interval, say, from t0 to t0 + T, where T is the
period of the fundamental component. This assumption enables the waveform
to be expanded by either Fourier or Walsh series. The fundamental component
is then extracted and then used to calculate either the impedance to the fault or
differential current quantities.
In the case of the Fourier transform method, no assumption as to the nature
of faulted waveform is necessary. Both the voltage and current waveforms
within the data window are transformed to the frequency domain. These
transformed quantities are then used to calculate the apparent impedance to the
fault.

5.2 Fourier-analysis-based algorithms

It was shown in Section 2.6.1 that any function of time (f(t)) can be represented
by a Fourier series and each coefficient of the series can be found according to
the formula given in eqns. 2.60—2.62.
Voltage and current waveforms are of course functions of time and they can
be consequently expanded using the Fourier series. If we take, for example, a
voltage waveform v(t), then
00 00

v[t)=-r+ > an cos na)ot+ > bn sin n(Od (5.1)

and from eqns. 2.61 and 2.62

2 f%T
a = v(t)
S{ cos na)ot dt, n = 0, 1, . . . (5.2)
" T\
J to
 Fourier analysis and Walsh function based techniques 73

2 .
v(t) sin na)ot dt, n = 1, 2, . . . (5.3)

where a)0 is the angular frequency of the fundamental component and T is its
period.
Eqns. 5.2 and 5.3 show that the fundamental component of a voltage and/or
current waveform can be extracted from the corresponding faulted waveform
simply by setting n = 1.

5.2.1 The full cycle window algorithm [1-3]

Basic approach
The basic approach used in this algorithm is to extract the fundamental
component of a waveform by correlating one cycle of the faulted waveform with
stored reference sine and cosine waves, as outlined in the previous Section.

Derivation of the algorithm

Let Vx and Vy be the real and imaginary parts, respectively, of the phasor that
represents the fundamental component of the faulted voltage waveform v(t). If
the time under consideration is t0, then Vx can be found from eqn. 5.2 such that

v(t) cos (o0t dt (5.4)

Now let N be the number of samples per cycle of the fundamental component,
At the sampling time interval, tj=jAt the time of thejth sample and T=NAt
the period of the fundamental component.
The integral that appears in eqn. 5.4 can then be evaluated using the
rectangular method, which leads to

c o s l 4 c o s
^ * • ' ' v^j)

COS (DolN-i+vit/t) COS (O0tN]At

2 ^ /2nj\ 2 "
Vx^% Vj cos I —)=^WxJVj (5.5)
7=0 \ / j=o

where »,-= »((;) IS thejth sample of the voltage waveform, and Wxj is the
weighting factor of thejth sample used to calculate Vn which in turn is given by
2n
WxJ = cos (o0tj = cos — j At

or, in the alternative form of

j = 0, 1, . . . N (5.6)
74 Fourier analysis and Walsh function based techniques

sin cot —

multiplier averager
sampler

cos ©t -

Figure 5.1 Schematic representation of the full-cycle window Fourier algorithm

Similarly, it is possible to approximate Vy in terms of discrete waveform samples

and their weighting factors as follows:
9 9
(5.7)

where Wvj is the weighting factor of the jth sample used to calculate Fv, as
defined in eqn. 5.8
WyJ = sin(2Jtj/N), 7 = 0, 1, . . . JV (5.8)
The algorithm described by eqns. 5.5 to 5.8 can also be used to calculate the
real and imaginary parts Ix and ly of the fundamental current component from
current waveform samples, and Figure 5.1 shows a schematic representation of
the algorithm.
The weighting factors defined by eqns. 5.6 and 5.8 can be made to have
values of 0, ± 1 , ± 1 / 2 , and ± V 3 / 2 , which in turn are suitable for computer
application, if the sampling rate is suitably chosen. For example, h*W= 12, then
the values of Wxj and WyJ are as given in Table 5.1. This technique reduces the

Table 5.1 Weighting factors for iV= 12 samples per fundamental cycle

i /

1 0, 12 0 0, 6, 12
V3/2 1, 11 1/2 1,5
1/2 2, 10 V3/2 2.4
0 3, 9 1 3
-V3/2 4,8 -1/2 7, 11
-1/2 5.7 -V3/2 8, 10
-1 6 -1 9
 Fourier analysis and Walsh function based techniques 75

output

Figure 5.2 Frequency response of the full-cycle window Fourier algorithm

amount of computation involved in applying the method in the real-time mode

of operation. Lower sampling rates, such as four or eight per cycle, can be used
where the application permits, but for many transmission-line applications
N= 12 is most appropriate.
When the weighting factors of Table 5.1 are used, eqns. 5.5 and 5.7 take the
following simplified form:
Vs
o I 2, (5.9)

VJJ(V-V+L{V-
V3 ]
^10 ) I 3.10)

Figure 5.2 represents the frequency response of a full-cycle window algorithm

from which DC and/or harmonic components are completely filtered out. This
is a useful and important property because it renders subsequent calculations,
e.g. circuit impedances, completely immune to error due to discrete harmonic
distortion of the signal measurands.

5.2.2 Fractional-cycle window algorithms

5.2.2.1 Half-cycle window algorithm [4]
This algorithm is based on correlating the fault waveforms with sine and cosine
functions having a frequency equal to that of the fundamental component of the
waveform. It is therefore basically the same as that of the full-cycle window,
except that it uses information corresponding to only one half cycle.
76 Fourier analysis and Walsh function based techniques

Let Fx 1/2 and Vyt in be the real and imaginary parts of the phasor that
represents the fundamental component derived from a half-cycle window.
Thus, by applying eqns. 5.2 and 5.3 to a half-cycle window, we obtain
po+772

I *o '*
sin a)ot dt (5.12)
(772) I v{t) si

Following the same procedure as that followed in Section 5.2.1, it is possible to

express Vx 1/2 and Vyj 1/2 in terms of waveform samples Sj9 and their correspond-
ing weighting factor WxJ and WyJy (j= 1, 2, . . . , N/2) such that
4 N/2
y X
x, 1/2 ~ *T 2.J J J \0.l3)

4 N/2

7=1

If N= 12, then the waveform coefficients can be written in the simplified form
If V3
V3 1
(5.15)

i r i V3 i
^,,/2 = 3 ^3 + 2 ("i + " s ) + — ("J + PO (5.16)

By comparing eqns. 5.5 and 5.7 with eqns. 5.13 and 5.14, or eqns. 5.9 and 5.10
with 5.15 and 5.16, it can be seen that the half-cycle window algorithm uses
only half the number of samples of those used in the full-cycle algorithm.
However, although this algorithm appears to be faster than that of the
alternative it has the disadvantage of introducing error, specifically due to any
aperiodic component and/or even harmonics. This can be seen in Figure 5.3
which represents the frequency response of the half-cycle-window algorithm.
The useful property of a rejection of odd harmonics is however retained.

5.2.2.2 Sub-cycle window algorithm [5, 6]

In an attempt to reduce the error introduced into half-cycle windows in the
presence of aperiodic components, the sub-cycle window algorithm correlates
faulted waveforms with sine/cosine functions having a period equal to the data
window length.
In this algorithm, the current and voltage waveforms are assumed to have the
following mathematical form.
i(t) = 70 e-'/T« + /, cos(ftV ~ 0/i) + h e~'/r> cos(«y - y) (5.17)
 Fourier analysis and Walsh function based techniques 11

output
input 1.0

0.75

0.5

0.25

3
CO

Figure 5.3 Frequency response of the half-cycle window Fourier algorithm

v(t)=Voc-"'o+Vl (5.18)

where

/0, Vo are the amplitudes of any decaying DC current and voltage components
I\y Vx are the amplitudes of the fundamental components of current and
voltage
Ip, Vp are the amplitudes of any assumed decaying current and voltage
oscillations
r0, Xp are the decaying time constants for any DC and high frequency
oscillations and
O)0, (Op are the frequencies of the fundamental and any high-frequency oscilla-
tions.

Since the data processed falls within a window (of length 7^,) it is convenient to
take the middle of the window as a reference. This can be achieved by shifting
the time scale / t o a new variable /', as shown in Figure 5.4, such that

(5.19)

where

If V'x and Vy are the quantities resulting from the correlation of the shifted
78 Fourier analysis and Walsh function based techniques

I*
2

Figure 5.4 Shifting of the time scale t to the variable t' where t'^t—fh +Tw/2).
(a) = original waveform, (b) = shifted version
* purely sinusoidal waveform is assumed

waveform v(t') with cosine and sine functions having a period equal to the time
span of data window Tw, then

V =I v(t') cos wj' dt' (5.20)

x
J-,...
V;= v(tf) sin coj'dt' (5.21)

where

is the frequency of correlating sine/cosine function.

 Fourier analysis and Walsh function based techniques 79

The amplitude Vx and phase angle 6VI of the fundamental component of v(t)
can be determined from V'x9 V'y by using the following equations:

(5.22)

The factors k and p depend on data window length Tm and they can be
determined as follows:

r sin nr
(5.23)

where r = a)Ja)0.
Although this algorithm reduces the error due to any aperiodic components,
it causes the error due to high-frequency oscillations to increase. From a
practical point of view, this algorithm is therefore most suitable for applications
where oscillatory components in the currents and voltages are relatively small,
e.g. in distribution system applications.
5.2.3 Fourier-transform-based algorithm [7]
Basic approach
This technique does not impose any assumption as to the nature of the relaying
waveform within any chosen window of information analysed. The basic
approach involves applying the Fourier transform to a data window within the
voltage and current waveforms. The impedance is then calculated from the
voltage and current transforms. By progressively advancing the window, it is
possible to observe the impedance variation with time after a fault, which in
turn is used to determine whether there is a fault within the protected zone.

Outline of the algorithm

Consider the assumed current and voltage waveforms shown in Figure 5.5,
where the window is shown to be from T, to T2. Since we are interested in
finding the Fourier transform of this window of information, we shall concen-
trate our attention on the voltage and current quantities therein. Let us assume
that the voltage waveform v(t) is made up of three components, such that
v(t) = vi(t) + v2(t) + vs(t) (5.24)
with

v(l) T^t^Tt (^26)

80 Fourier analysis and Walsh function based techniques

> • time

A Vi a

/ ^
y\ f V\
time
\ \
1 Ti w \
fault
inception V
v

Figure 5.5 (a) Faulted voltage and current waveform (b) Decomposition of voltage
waveform v into three components vi9 z*j, v3

t>T2
(5.27)
1 t<T2

According to the principles of circuit theory, each of the voltages »,, v2 and vz
would produce a corresponding current according to the following equation:

k=l, 2, 3 (5.28)

where Z(p) is the operational value of the circuit impedance.

By the principle of superposition, the three components add to give the total
current i(/), such that

(5.29)
 Fourier analysis and Walsh function based techniques 81

Now the Fourier transform of the window of voltage and current, v2(t) and i2(t),
can be calculated using eqn. 2.73 to obtain

»=["
v2(ja>) =
J -oo
v2(t)e-jm'dt (5.30)

The limits defined in eqn. 5.26 allow the Fourier transform of the given window
to be reduced:

1 = 1 v(t)e-ja)tdt (5.31)
J T
The Fourier transform of the corresponding current i2(t) is given by

-L i2(t)e-J°"dt

It will be noted that v2(j(o) can be easily evaluated using numerical methods
(5.32)

(such as the trapezoidal technique), because v2{t) is readily available. However,

this is not the case for i2(j<i>) because i2(t) is not readily available.
Before proceeding to a derivation ofi2(j(o), it is useful to note the following
points:
(i) According to the principles of circuit theory [8], when the voltage
collapses, the current in the circuit does not cease immediately. Therefore,
if, for example, voltage z>, collapses at / = Tn the current ix(t) which flows
for time t > Tc is approximately
it(t) = i(Tc)e'{"T')/r t>Tc (5.33)
where r is the time constant of the circuit.
(ii) The current that would flow within the data window consists of two
components, the first being due to the voltage component v{(t) (as
explained in (i) above), and the second due to the voltage component v2(t).
(iii) By the same reasoning as given in (i) above, the current i2(t) is not confined
to the window, i.e. it does not cease at time T2, but rather continues to flow
thereafter, despite the collapse of voltage v2 at T2.
Now if the total current within the window is i(t)> then
, (5.34)
But, according to eqn. 5.33, i{(t) at / > Tx would be given by substituting Tc by

j (5.35)
By substituting this equation into eqn. 5.34, we obtain
h(t) = M-i(Tl)cru-T>)'t+/„(!) 7-,«.'<7i (5.36)
It must be emphasised that this equation is valid only during the window
bounded by 7*, =^ / ^ T2.
82 Fourier analysis and Walsh function based techniques

At / = T2 the voltage component v2(t) collapses and therefore i2(t) continues to

flow according to eqn. 5.33 (see remarks (i) and (iii)):
h(t)=h(T2)e-«-T^+fh2(t) t>T2 (5.37)
where i2(T2) is t n e value of the current i2(t) at / = T2, and can be found from eqn.
5.36 by substituting / by T2.
i2(T2) = i(T2)-i(Tl)c-{T^)/t+MT2) (5.38)
By combining eqns. 5.37 and 5.38, we obtain the current which flows in
response to the application of the window bounded by times Tu T2:
h(t) = [i(T2)~i(Tt)c-^-T'^+MT2)]c-"-T^ +fK(t) t>T2 (5.39)
The components fhx{t) and fh2(t) are added to take into account any high-
frequency components caused by travelling waves. However, practically, for
lines of length not more than 250 km, these high-frequency components can be
neglected because their magnitudes are small compared with the components at
or near power frequency. Therefore, by substituting i2(t) described by eqns. 5.36
and 5.39 into eqn. 5.32, and neglecting^(t) and^ 2 W> w e obtain:

= i2(t)e-j(otdt= I 2 i2(t)e'jatdt+ I i
J TX J r, J T2
or
fr2
= [i(0 - KTX )e" ( / - r ' ) / r ] c-^dt

J Ti

which results in J T2
l T
^ ^ (5.40)

where

= i{t)z-jwtdt (5.41)

Computational aspects
Since voltage and current waveforms are presented to the algorithm in the form
of samples, the integration terms that appear in v2(ja)) and i2{j(o) should
therefore be evaluated numerically.
Let At be the sampling interval. Then

T2=TX + TW
= (N+n)At
 Fourier analysis and Walsh function based techniques 83

where n is an integer which defines the position of the data window, and Nis the
number of sampling intervals within the data window of length Tw.
By substituting Tu Tw and T2 (defined in eqn. 5.42) into eqn. 5.31, we obtain

= v(t)e*"4t
Jn
where v2n(jco) is the transform of v2(t) for a window starting at Tx = n&t. When
the trapezoidal rule is applied to this equation, we obtain

1
•v[(n + N-

or
_ JV— 1 -i

j(o)=— Wnvn+ W , + * w + 2 2Wn+mvn+m (5.43)

where vn is the nth voltage sample, and Wn+m is the weighting factor of the
(#-fm)th voltage sample.
When the window is moved by one sample, the transform of the voltage v2(t)
within the new window, i.e. v2(n+^(ja)), can be found recursively by simply
dropping the oldest sample vn and adding the newest sample vn+N+\9 such that

H <«+oO) = V2n(j<*>) + 0* (5.44)

where

At
Q«= —7T [ - W > « - W^+i»«+i + Wm+NOH+N+ Wn+N+lvn+N+l]

This equation can be verified mathematically by following the foregoing

procedure and substituting for 7i = (n+ I)At and T 2 = (n + iV+ 1)A/. The
resulting equation is then reduced to the form described by eqn. 5.44. Thus, to
update the voltage transform as the window advances requires only the addition
of the sampled function Q,, to the integral approximation taken over the period
immediately before the one in question.
Computational efficiency can be improved if the extraction frequency a) is so
chosen so that the period of Fourier kernel terms (e~-^"A/) corresponds to an
integer multiple k of the sampling interval, i.e., 2Jt/(O = kAt. If this is done, the
weighting coefficients Wn+m can be pre-evaluated and used in a cyclic process as
the window moves, thus obviating the necessity for an online evaluation of the
84 Fourier analysis and Walsh function based techniques

Fourier kernels. The previous method can be used to evaluate the current
transform i2(Ja)) defined by eqns. 5.40 and 5.41, which in turn results in

A*r ^ 1

where Mn= Wn+Nin+N~ Wjn, in = nth current sample, i2n(Jo)) = the transform of
i2(t) relating to the window which starts at T—

5.3 Walsh-function-based algorithms

Digital relaying using Walsh functions was originally developed to reduce the
burden on computers by using only add and shift operations in fixed point
format, thereby avoiding the multiply, divide, square and square-root ope-
rations. The computational simplification is made possible by the fact that
Walsh functions assume values of only ± 1.

5.3.1 Basic principles

The basic principles of digital protection using Walsh function techniques
consists of expanding the faulted waveforms of voltage and current by using
Walsh series, from which Walsh coefficients {Wh k= 1, 2, . . .} are determined.
These functions are then used either indirectly, through the Fourier-Walsh
relation, or directly to determine the amplitude and phase angle of the
fundamental components of the voltage and current waveforms, which are then
used to determine the impedance to the fault.

5.3.2 Development of basic algorithm

It will be seen from Sections 2.6 and 2.7 that any assumed periodic function can
be expanded using either the Fourier series or the Walsh series. Therefore, if for
example a faulted voltage waveform v(t) is assumed to be periodic in the
interval from tx to tx + T (assuming /i = 0), then, from eqns. 2.83 and 2.84 we
obtain
00

v(t) = FQ + ^ (V2F 2 n _, sin na)^ + V2F 2 n cos na>ot) (5.46)

with
1 [T
FO=T\ v(t)dt
Jo
V2 fT
F2n-X=— v(t) sin na)0t dt (5.47)
Jo
 Fourier analysis and Walsh function based techniques 85

v(t) cos no)d dt

where F o is the DC component and F2n, F2n+1 are the RMS values of the real and
imaginary components of the nth harmonic, respectively. As indicated previous-
ly, it is the Fourier series component values at fundamental (or power)
frequency that are used to determine the measured impedance. However, it is
relatively easy to determine these by first evaluating the Walsh component
values. In this respect, it will be recalled that by using the Walsh series, v(t) can
be approximated using eqns. 2.85 to obtain

^) (5.48)

with the kth Walsh coefficient (Wk) given by

Wk=X~ fT v(t) Waif *, j\dt k = 0, 1, 2, . . . (5.49)

Fourier and Walsh coefficients are interrelated by matrix A, as described by

eqn. 2.92 which is rewritten below:
F=ATW (5.50)
The Fourier to Walsh transformation matrix A and its transpose consist of fixed
constant values (see eqn. 2.90) and, given a determination of Walsh compo-
nents using eqn. 5.49, it is relatively straightforward to determine the funda-
mental or where required harmonic Fourier series components using eqn. 5.50
above.

533 Algorithm for Walsh function determination [9]

It is possible to reduce the amount of real-time computation by employing a fast
algori#im, which is used to calculate the Walsh function coefficients Wk for any
arbitrary waveform. Consider a voltage waveform that is available in the form
of samples at a rate of TV samples per cycle. In this case, the sampling interval A*
is related to the time over which the signal is periodic (7) and the number of
samples therein (N) by T=NAt. In this case tj=jAt, j = 0, 1, 2, . . . , N and
v(tj) = v(j&t). Then

Wal(*, /j) = Wal( * / — J = Wal(*, jA/'),

where
At
86 Fourier analysis and Walsh function based techniques

When this is done, the Walsh coefficients Wk can be calculated according to eqn.
5.49, such that

v(t) • Wal(£, t')dt

The integral on the right hand side of the above equation can usefully be
evaluated using the trapezoidal rule, which results in

Wal(A,

1
+ v(tN-}) • Wal(*, (N-\)Atf)-¥-v(tN) • Wal(t, JVA*') A* (5.51)

Now let v(tj) = vj, wj(k) = Wal(A, jA/') and fj = v(tj) • Wal(*JA/') = »,• • i^.(A), so
that by substituting into eqn. 5.51 we obtain

(5.52)

Walsh coefficients Wk (k= 1, 2, . . ) are evaluated as each new sample of v(t) is

taken. This can be done by dropping off v0 (the oldest value) and adding or
subtracting % + I (the newest value). It is possible to imagine the waveform v(t)
as passing from right to left through a window which contains the Walsh
components. Figure 5.6 shows the effect of movement from right to left of a
sampled sinusoidal waveform v(t) through a window which contains the Walsh
function Wal(l, tf). The effect of this movement on resulting f(jAt) is also
shown.
Generally, if the voltage waveform moves an equivalent of sAt with respect to
the initial position of sample vp then the sample which occupies the position of Vj
will be vj+s. This simply means that moving v(t) s steps leftwards is equivalent to
a time displacement of sAt. This in turn results in
fj+s = vj+s-wj(k) .=0,1,2,... (5.53)
Therefore, by moving the waveform v(t) by s • At, the Walsh coefficient Wk(s) is
given as

=]faf,+ji++, , + • . .J+1
Wt{s) =]faf,+ji +1 + • • . J|-1+
1+,, +^
+^ jf+, | (5.54)

If for example N=8, then either with the help of Figure 2.6 or by using the
discrete representation of the Walsh function given in Section 2.7.3, it can be
seen that
in i i
o(s) = - - * , + * , + , + . . . v7+s + - vB+s (5.55)
 Fourier analysis and Walsh function based techniques 87

Wal (1 + f)

vft + 1)

f A
xDF

Figure 5.6 The mechanism of movementfrom right to left of a waveform through a window
that contains W a l ( l , t)
(a) Walsh function window which contains Wal(l, t')
(b) Sampled sinusoidal waveform Vj = v(tj) = v(jAt)
(c) Resulting function f(jM) = v(tj) • Wal(l, /')
(d) The function v(tj+l) = v((j+\)At) resulting from v(tj) after being
moved to the left by one sampling interval
(e) The resulting^, = / ( ( / ' + 1)A/) = »((/+ l)A*) ' W a l ( l , t')
88 Fourier analysis and Walsh function based techniques

Therefore W0(s + 1) can be found recursively by dropping off the oldest sample
and adding the newest sample v9+s to obtain
ir i i i i
-\ --v,--o1+, + -vli+, + -v9+,

Similarly, Wx can be found recursively as

If 1 1 1 1 1
Wx(s+ 1) = Wx{s) + H - - vs-~ vl+s + vt+s + v5+s-- v8+s-- v9+s

J = 0, 1,2, . . . (5.57)
From a computational point of view, not much is gained from calculating W2
recursively since it can be calculated directly as
lfl 1 1
W2(s)=~\-vs + vl+s--v3+s--vt+s-v5+s + v7+s + ~vs+s\ (5.58)

The remaining coefficients can be found in exactly the same way and are given
in eqn. 5.59 below.
in i

in I i
- 1 - vs. - v2+1 + o 4+J + - vs+s

iri i I
(5.59)
j j b , - V2+, + v6+1 - - vs+I

lfl 1 "1
- " ! , - Vi+S + ~ VS+S
lfl

It will be noted that, except for the end-points of the interval (0, 7), the
application of the trapezoidal rule leads to the cancellation of the sample value
where a jump occurs in the appropriate Walsh function. This is because the
area of the function f(tj) = v(tj) • Wal(A:, /j) around any such jump has equal
parts but with opposite sign (see Figure 5.6*).

53 A Estimation of the amplitude and phase angle of fundamental components

5.3.4.1 Indirect method
This method involves the determination of the amplitude and phase angle of the
fundamental component of a faulted voltage or current waveform using Walsh
coefficients and the Walsh—Fourier transformation matrix A. Let Wk%n k =
 Fourier analysis and Walsh function based techniaues 8Q

0, 1, 2, . . . be the Walsh coefficients of a faulted voltage waveform, determined

as discussed in the previous subsection. Now, using these coefficients, it is
possible to estimate the Fourier coefficients FUv and F 2 , w which are related to
the fundamental components. If we consider a 10 X 10i4-matrix, then, by eqn.
5.50, F, v and F2t v can be expressed in terms of WxtV as follows:
F, , = 0.9 Wx , - 0 . 3 7 3 W5 ,-0.074 W9,
(5.60)
F 2 ,, = 0.9 W2t, + 0.373 W^v-0.074 Wl0,v
It is then relatively easy to calculate the amplitude and phase angle of the
fundamental Fourier series component from eqn. 5.61.
)
= V2 y/Fl. + ft.
and
fl.1 = tan- I (F 2 .,/F 1 ,.)
The amplitude and phase angle of the fundamental component of the current
waveform can likewise be estimated in exactly the same way.

53A.2 Direct method [9]

This method was developed principally to avoid the square root and squaring
operations necessary when using previous methods. It also involves using
Walsh coefficients directly (i.e. without going into a determination of the
Fourier coefficient) to estimate the fundamental amplitude and its phase angle;
this is based on two theorems.
The first theorem states that 'the maximum value of a sinusoid and the time
of its occurrence can be estimated in terms of the first and second Walsh
coefficients for the sinusoid'.
Assume g\ (/) is a sinusoidal waveform with interval T, which is divided into
16 sub-intervals such that A / = 7V16. Ifg^t) is expressed within Tin terms of
Walsh functions, such that
10

=2 (5.62)

then
(i) to an accuracy of (+4.28% to —4.8%) the maximum value of g{{t)
(GM) is given as Gmax= 1.0822(|W,| + \W2\) + 0AU\\Wt\ -\W2\\ (5.63)
(ii) the maximum value ofgi(/) occurs in the mth sub-interval, within which the
following formulas are satisfied:

Wal(3, mAt)=S(W2) . S(\W2\ - \W{\) (5.64)

Wal(10,
90 Fourier analysis and Walsh function based techniques

where £(#) = sign(x) is a function which is positive when x is positive, and

negative when x is negative. For x = 0 both plus and minus signs are to be used,
and the maximum is not unique.
The second theorem states that the accuracy of estimating the amplitude of a
sinusoid and its time of occurrence can both be improved if all 16 Walsh
functions with unit interval are taken into account. Therefore, according to this
second theorem, if gi(t) is a given sinusoidal function with periodic time T,
which is divided into 16 sub-intervals, and i£g\(t) is expanded in terms of the 16
Walsh functions defined by eqn. 5.65,

/ \
Wall*,-I (5.65)

then

(i) to an accuracy of ±2.6%, the maximum value of g\(t), Gmax, can be

expressed in terms of Wx and W2 by the following formulas:

(ii) the maximum occurs in the mth sub-interval which satisfies the following:

Wai(2, mAl)=S(W2)

• 5(A)

Wal(10, mM')=
[S(W.2)

where

a = 0.414 y = 0.198 and B = 0.0823

 Fourier analysis and Walsh function based techniques 91

53.5 Determination of Walsh coefficients for pure sinusoidal waveforms

To apply the previous theorems, one must first determine the Walsh coefficients
associated with the fundamental component of the assumed sinusoidal wave-
form. Consider a voltage waveform, and let its fundamental component vx{t) be
such that

Vx (/) = V2Flv cos a)ot + V2F 2v sin co0t (5.66)

To avoid any confusion that may arise between the Walsh coefficients
ftfk (£ = 0, 1, . . .) for the whole voltage waveform and those that relate only to
the fundamental component of that waveform, we shall denote the Walsh
coefficient due to the fundamental by Wxk (k = 0, 1, . . .). It follows from eqn.
5.62 that the fundamental component can be expressed as

(5.67)

However, the coefficients Wn> v and Wxx v c a n D e found from their corresponding
Fourier coefficients, using the inverse of eqn. 5.50 as

W
"-~0*F- ,5.68)
^ 09F

This is so because F t = 0 for all i except i = 1 and i = 2 (see eqn. 5.46). On the
other hand, Flv and F2v can be expressed in terms of the Walsh coefficients due
to the whole faulted waveform. Assuming a 10 X 10 A matrix, Fu and F2v can be
found according to eqn. 5.60. By substituting the latter equations into eqn. 5.68,
we obtain

Wn , = 0.81 Wx ,-0.3357 W5 ,-0.0666 W9 v

(5.69)
WI2t, = 0.81 W%v + 0.3357 W^v-0.0666 Wl0,9

It will be noted that Wn^v and PVl2v are equal to Wx and W2 stated in the
amplitude and phase theorems.

5.4 References
1 SLEMON, G.R., ROBERTSON, S.D.T., and RAMAMOORTY, M.: 'High speed
protection of power systems based on improved power systems models', CIGRE,
Paris, June 1986, Paper 31-09
2 RAMAMOORTY, M.: 'Application of digital computers to power system protection',
J. InsL Eng. India, 1972, 52, pp. 235-238
3 McLAREN, P.C, and REDFERN, M.A.: 'Fourier-series techniques applied to
distance protection', Proc. 1EE, 1975, 122, pp. 1301-1305
92 Fourier analysis and Walsh function based techniques

4 PHADKE, A.G., HIBKA, T., and IBRAHIM, M.A.: 'A digital computer system for
EHV substations: analysis and field test', IEEE Trans. 1976, PAS-95, pp. 291-301
5 WISZNIEWSKI, A.: 'Signal recognition in protective relaying' in 'Developments in
power system protection,' IEE Conf. Publ. 185, 1980, pp. 132-136
6 WISZNIEWSKI, A.: 'How to reduce errors of distance fault locating algorithms',
IEEE Trans., 1981, PAS-100, pp. 4815-4820
7 JOHNS, A.T., and MARTIN, M.A.: 'Fundamental digital approach to the distance
protection of EHV transmission lines', Proc. IEE 1978, 125, pp. 377-384
8 DESOER, C.A., and KUH, E.S.: 'Basic circuit theory', (McGraw-Hill, 1969)
9 HORTON, J.W.: 'The use of Walsh functions for high-speed digital relaying', IEEE
PES Summer Meeting, 1975, July 20-25, Paper A 75 582-7
 Chapter 6
Least squares based methods

6.1 Introduction
In this Chapter we shall discuss techniques used to fit faulted current and
voltage waveforms, each to a sinusoidal waveform containing a fundamental
component, a decaying/constant DC component and/or harmonics.
These techniques use the least squares (LSQ) method to minimise the fitting
error, and all have the common goal of extracting the fundamental components
of voltage and current waveforms, to calculate the impedance to the fault or the
comparison of current-based signals in digital differential protection.

6.2 Integral LSQ fit [1]

6.2.1 Basic assumptions
It is assumed that the monitored signal, a voltage waveform v(t), is approxi-
mated by a function g(t), which consists of a fundamental component, an
exponentially decaying DC component, and/or harmonics. Therefore, in
general terms, g(t) can be written as
M

g(t) = K, e~'/r + ]T (K2m cos ma)0t + K2m+, sin ma)Qt) (6.1)

m=l

where Ku K2, . . ., K2M+i are the unknown constants required to be determined,

M is the highest harmonic considered, r is the time constant of the decaying DC
component and (o0 is the angular frequency of the fundamental component.
6.2.2 Determination of unknown coefficients
The unknown constants of eqn. 6.1 can be determined using the least squares
method explained in Section 2.5.1. Consider a voltage waveform v{t). Now the
integral of the squared errors S is, as previously explained, given by eqn. 6.2, in
which g{t) is the approximation defined in eqn. 6.1:

=l[v{t)- g{t)Vdl (6.2)

When eqns. 6.1 and 6.2 are combined together we get

M

ma)ot)fdt (6.3)
J
94 Least squares based methods

As discussed in Chapter 2, the best fit occurs when S is a minimum, and this is
satisfied if
dS
— =0 (i=\,2,...,2M+\) (6.4)

By performing the partial derivative with respect to Ku K2m and K2m+U we

obtain

dS

cos sin ma)ot)] €~ tlxdt

dS f
~dK.- =0 = - 2 [v{t)-Ki e'"T
•m J (6.5)
cos m+, sin ma)ot)] cos m(O0t dt

dS f
=0=-2 [v{t)-Kx^
&2m+l J

cos ma)ot + K2m+i sin ma)ot)] sin ma)ot dt

The limits of integration in the previous equations are taken from /1 to t{ + !F(i.e.
over one fundamental period). After simplification, eqn. 6.5 then reduces to
f/, + 7- ftt + T
K{ e" 2 / / V/= v(t) etlxdt
J /, J /,
2 f'.+T-
K2m = j , I »(0 cos mft>0^ dt (6.6)
J t\

2 f'l+r
K2m+t= -j, I »(0 sin ma)ot dt
J i.
These equations have been derived from eqn. 6.5 using the following identities:

e sin mx dx = - [a sin mx — m cos

V2 1
cos
eax cos mx dx = — 5 ia mx — m sin mx]eax]x\+2 = 0
a -¥m
1

It must be remembered that the monitored waveform v(t) is available in the

form of discrete samples.
 Least squares based methods 95

Derivation of coefficients in terms of measured waveform samples

If the decaying DC time constant r is known, then it is possible to express Kx in
terms of the measured samples Sn = v(tn), (n= 1, 2, . . . , N, where N is the
number of samples per cycle). Let A/ = sampling interval; then tn=(n— \)&t,
which gives tx = 0, and T= NAt is the fundamental period time.
By approximating the integral using the trapezoidal method, therefore, eqn.
6.6 can be written as

- KJM = -[v{h) e-""+2v(t2) e-*/T+ • • • +2o(fo-.) e -'*- /T + s(M ^/T]At

or
AT— I

K} = WnSx + WINSN+ 2 2WlJS, (6.7)

»= 2

where

and WXn is the weighting factor of the ith sample used to calculate Ku which is
equal to e'^/I.
If the trapezoidal method is applied to the second eqn. 6.6, we obtain
fl (2n\ (2n\
b f t ) cos ml YJ t\ + v(h) cos ml — I / 2 +•
/2n\ 1 /2JT\ 1
cos m + V C S m 1N
*JV-I) \~f) ^-i o ^ ° \~Tl
or N l
i r ~ i
L «=2 J
where
W2my n ~ weighting factor for nth sample for calculating K2m
(2n\ (2n{n-\)^
= cos m

Similarly K2m+i can be expressed as

1 T JV-I -I
(6.9)
L «=2 J
where
W/2m+i,B = weighting factor of wth sample used to calculate K2m+\
2JZ\ . [2n(n-\)
= sin ml — nl/n = sin ml
T/ \ N
96 Least squares based methods

6.2.3 Implementation of the algorithm

The above algorithm can be implemented either to extract the fundamental

frequency component (as in the case of transmission line protection), or to
extract the second-harmonic component as well as the fundamental component.
The latter is particularly important when dealing with protection of trans-
formers, as will be discussed in detail in Chapter 10.

6.3 Power series LSQ fit [2]

6.3.1 Basic assumptions

This algorithm assumes that the current and voltage waveforms contain only
constant DC and fundamental components. Taking the voltage signal v(t) as an
example, then v(t) is assumed to take the form
(6.10)

6.3.2 Shifted waveform

To minimise the computational burden, the fit is performed using a conti-

nuously shifted time reference frame. This may be done by noting that if/* is
the shifted time reference frame and ts is the time shift, then the new time
reference frame t* is defined as
/• = / - / , (6.11)
If the shifted time reference frame is set to correspond to half the number of
sampling intervals (N— l)A//2, then

Now consider the samples 5,(i = 1, 2, . . . , N) occurring at times

(JV-1)
t-(N-l)At, t-(N-2)At, ..., I — At, . . . (6.13)

When these times are transformed to the new reference frame t*, using eqns.
6.11 and 6.12, the sequence 6.13 becomes
N- \\ (N-1
0 >
{ ) A ' • • • ' ~ A t
' • •'

)(J (6.14)
 Least squares based methods 97

origin of the shifted

time reference

A 2A

Figure 6.1 Sampled data plotted against a shifted time reference

This sequence shows that the sampling time in the new reference t*, is
independent of time. Figure 6.1 explains the meaning of the new time reference
in association with an assumed sinusoidal waveform.
By combining eqns. 6.10 and 6.11, we then obtain
sin[wo{t* +
or
v(t) = Kl + K2 sin(a)ot* (6.15)
where

6.3.3 Approximating the shifted waveform by a power series

When eqn. 6.16 is expanded, we obtain
P* = p(/*) = KX + K2 sin a>ot* cos /3* + K2 cos (o0t* sin
But

sin
3!
and
(«>ot*)2
cos a)ot*
2!
Therefore the first equation can be expressed as
v * = a0 + axx + a2x2 (6.16)
98 Least squares based methods

where
x = coQt*
ao = K\ + K2 sin #! = K2 COS /?*
a3=(-K2 cos
Eqn. 6.16 expresses the sampled value in terms of ^-coefficients. In other words,
the monitored waveform, which is assumed to consist of a constant term and a
fundamental component, is fitted by a power series of third order. The
amplitude K2 and phase angle (i of the fundamental component can be
expressed in terms of coefficients ax and % as follows:

(6.17)

= tan

Determination of the power series coefficients

The ^-coefficients of eqn. 6.16 can be determined using the least squares method
of Section 2.5.1. If, for example, the measured voltage waveform samples are
assumed to be ^(i= 1, 2, . . . , N), the sum of the squares of error S can be
found using eqns. 2.37 and 2.38 ifj>, is replaced by v( and u by v*, such that

(6.18)

According to eqn. 2.39, the summation of the square of the error S is a minimum
if
dS

When this is applied to eqn. 6.18, we obtain the following set of equations:

2 " 2*? 2*7 2*?

(6.19)

2*7 2*7 2*? 2*? 2**7

2*? 2*? 2*7 2*? 2^
 Least squares based methods 99

Note that the range of summation in eqn. 6.19 is from 1 to N and vh *,- are
evaluated at times corresponding to *,• = / — (N—i)At.
Note also that only ax and a2 are required to determine the amplitude and
phase angle of the fundamental component as per eqn. 6.17. It is, therefore,
sufficient to solve eqn. 6.19 only for a{ and a2. Furthermore, the computational
burden can be greatly reduced if the data are shifted to the new reference frame
t* (which causes x{ to remain constant) and N is chosen to be an odd number.
This results in all sums involving odd powers of x becoming zero. Therefore, by
manipulating eqn. 6.19, the constants a{ and a2 can then be expressed as follows:

a, = M2
(6.20)

a2:== J

where

M{=- I-"

(y A2-
The constants Aft, f = 1, . . . , 4 can be determined beforehand once the number
of points N, and the sampling interval, have been selected. It should be noted
that the limit of the summation is from £ = 1 to i = N.
100 Least squares based methods

6.4 Multi-variable series LSQ technique [3]

6.4.1 Basic assumptions

The development of the multivariable series based algorithm is based on the

following assumptions:
(i) In practice, even harmonics are not present in the faulted voltage and
current waveforms
(ii) Fifth and higher order harmonics are blocked from reaching the relay by
signal preconditioning equipment.

Therefore, it will be assumed that the voltage waveform v(t) consists of a

decaying DC component, the fundamental and the third harmonic component,
such that

v(t)=Kl et/j + K2 sin(a)ot + 0,) + K3 sin(3<yo/+ 03) (6.21)

6.4.2 Derivation of the multi-variable series

The exponential term €~ tl% of eqn. 6.21 can be expanded using Taylor's series,
such that
2 3
t I t I t
e //r==1 + (6 22)
" -;^?"ii? "- -
By considering only the first three terms of this expression, eqn. 6.21 can be
expressed as

/ t2
v(t) = *, - kx - + *i —$ + *2i s

or
t t2
v(i)=k{—kx - + A;, — 5 + (k2X cos 0,) sin cwo/-f (k2X sin 0i) cos a)Qt

s m cos
4- (k23 cos 0 3 ) sin 3a)ot+(k23 #3) 3tt>0/ (6.23)

where kx = iC h A:2i = iT2, A:23 = i^3.

Therefore, at t = ^ the instantaneous value of the signal can be expressed as

v(tx) = A;, — kx — 4- A:, r~^4- (A:2! cos 0,) sin a>0/, 4- (k2x sin 0|) cos

+ {k2i cos 03) sin 3ft>o/| + (k23 sin 83) cos 3<w0/
 Least squares based methods 101

This equation may be written in the more convenient form

St = v(tx ) = anXx + 0,2#2 + 013*3 + ^14^4 + #15*5 + «16*6 + a\l*7 (6.24)

where Sx is the sample measured at time tx. The coefficients in eqn. 6.24 are
related only to the time at which the samples are taken, and they take the form

aX5 =

The ^-values are functions of the unknowns, and are given by

xx = k x x 2 = k2X c o s 9X x 3 = k2X s i n 6X
#4 = k23 cos 63 x5 = k23 sin 63
x*=-kxlx x7 = kx/(2t)2
The sample S2 measured at time t —12 can likewise be expressed as

x-¥ a22x2 + a23x3 + a2ix4 + a25x5 + a26x6 + a21x7 (6.25)

where a 2 i = 1, a22 = sin((Dot2), a23 = co$(a)Qt2), . . ., etc.

As mentioned previously, the ^-coefficients are functions of time. Therefore if
£j is taken as a time reference and the voltage is sampled at preselected times,
then the values of the coefficients of eqns. 6.25 and 6.26 can be specified. The
values of the sampled voltages Sx and £2 are also known so that, by proceeding in
this manner, m equations can be generated from m samples of voltages. These
equations can be written in matrix form:
A X _ S (6.26)
mX7 7X l~~mX 1
The elements of matrix A depend on the time reference and the rate of sampling
and, if the number of samples m is taken to be greater than the number of
unknowns (i.e. greater than seven), the matrix A is a rectangular matrix which
has m rows and seven columns.
The solution of eqn. 6.26 can be found by using the concept of the pseudo-
inverse, discussed in Section 2.5.1. This involves premultiplying both sides of
eqn. 6.26 by the left pseudoinverse of A, which results in
X = A1 S (6.27)
7X1 7Xm mX 1
where A1 is the pseudoinverse of A and is defined as

After determining the unknown vector X from eqn. 6.27, the magnitude of the
fundamental component of voltage K2 and its phase 0x can be found in terms of
x2 and x3 (see eqns. 6.23, 6.24).
102 Least squares based methods

6.5 Determination of measured impedance estimates

The three algorithms covered in this Chapter share the common goal of
determining the impedance to the fault using the fundamental components of
current and voltage already determined by each of the individual algorithms.
According to the integral LSQ fit algorithm, the impedance Z is found as

Z (6.28)

Now K2y, K2/ are the real parts of voltage and current fundamental components,
respectively, and are determined using eqn. 6.8 (with m= 1). Likewise, K3V&nd
K3I are the imaginary parts of voltage and current fundamental components,
respectively, which are determined by using eqn. 6.9 (with m— 1).
The measured impedance can alternatively be determined using the power
series LSQ fit by means of eqn. 6.29:
K2VL (iV
Z=-— r (6.29)

In the case of the power series, K2v, K21 a r e the amplitude of the voltage and
current fundamental components, respectively, and j3v, /?/ are the phase of the
voltage and current fundamental components, respectively. The latter are
determined by eqn. 6.17.
Finally, a measured impedance can be found using the multivariable series as
{ \ j [ \ j w j w
( }
Re[/]+jIm[/] Xy+jx3l '
The values x2F, x3V are the real and imaginary parts of the fundamental of the
voltage waveform and are determined according to eqn. 6.27, together with the
corresponding values derived for the current (;%, x3f).

6.6 References
1 LUCKETT, R.G., MUNDAY, P.J., and MURRAY, B.E.: 'A substation based
computer for control and protection' Developments in power system protection, IEE
Conf. Publ. 125, London, March 1975, pp. 252-260
2 BROOKS, A.W., Jr: 'Distance relaying using least-squares estimates of voltage,
current and impedance', Proc. IEEE PICA Conf, 77CH 1131-2, PWR, May 1977,
pp. 394-402
3 SACHDEV, M.S., and BARIBEAU, M.A.: 'A new algorithm for digital impedance
relays', IEEE Trans., 1979, PAS-98, pp. 2232-2240
 Chapter 7
Differential equation based techniques

7.1 Introduction
In this Chapter no special assumption will be made with regard to the content
of faulted current and voltage waveforms. The fundamental approach, which is
common to all algorithms covered in this Chapter, is based on the fact that all
protected equipment can be normally represented by differential equations of
either first or second order. The methods are described by reference to
transmission-line protection, since it is in this application that they are mainly
used. However, the methods can easily be extended to other items of plant. For
the purposes of this Chapter we shall assume the line length is such that shunt
capacitance can either be neglected or can be lumped into a single 'equivalent'
value.

7.2 Representation of transmission lines with capacitance

neglected
Generally, three-phase transmission lines can be represented by a set of
differential equations expressed in terms of basic electrical parameters, namely
resistance, inductance and capacitance. Consider first an element of length Ax
along a three-phase line (Figure 7.1). If it is assumed that Rk, Lk are the
resistance and inductance per unit length of the A;th phase, Lkl is the mutual

-czb 'trnv-
Figure 7.1 Elemental length of distributed transmission line represented by series resistance
and self and mutual inductances
104 Differential equation based techniques

inductance per unit length between kth and /th phases and ik9 vk are the current
and voltage of kth. phase, then the voltage drop across an infinitesimal length dx
of phase 'a' is:

dva=(Radx)i,+ (LA) Jt + {LJx) -£+ (LJx) j (

dva ( d\ dib dic

— = [R. + L . - ) i. + L*— + L -
dx \ dt] dt dt

Similarly, the voltage-current relations for phases b and c would be

dvb dia I d\ dic

dvc dia

If the line is assumed to be ideally transposed, we have

R a = R b = R c = Rs
La = Lb = Lc = Ls (7.4)
r r j r r r
ab ~~ ^ac ~ ^ba ~~ ^bc ~~ ^ca "~ ^cb "~ ^m

where Rs, Ls are the series resistance and self inductance per unit length of each
phase, and Lm is the mutual inductance per unit length between any two phases.
By substituting eqn. 7.4 into eqns. 7.1-7.3, we obtain:
dva I d\ dib dic
s s %a m m
~dx~\ dt) Yt^ dt

dvb dia ( d\ dic

— = Lm —+ R* + Ls - \ib + Lm — (7.5)
m v ;
dx dt \ dt r dt

Now the self and mutual parameters are related to the zero and positive phase
sequence parameters as follows:

(7.6)

and
 Differential equation based techniques 105

where Rl9 Lx are positive phase sequence resistance and inductance, respecti-
vely, Lo is the zero sequence inductance and i0 is the zero sequence current.
By combining eqns. 7.5 and 7.6, we then obtain

di0
at

dvb I d\ di0
X
'+(LQ-Ll)— (7.7)

Eqns. 7.1-7.3, 7.5 and 7.7 can be used to calculate the voltage drop between the
fault point and the relay location for different fault types; each of these will be
considered in the following subsections.

7.2.1 Single-phase to ground fault

Assume a solid single-phase to ground fault occurs on phase V at a distance x
from the relay location. The instantaneous value of the voltage va, which is the
voltage of phase V at the relaying point, can be calculated using eqns. 7.1 for
untransposed lines and eqns. 7.5 or 7.7 for assumed ideally transposed lines.
Using the instantaneous values of the voltages, currents and the rate of
change of the currents, the voltage va can be obtained using eqn. 7.1 such that

or

^ (7.8)
at

where

and
106 Differential equation based techniques

Equations relating to transposed lines can be expressed in the same way, and in
this case eqns. 7.5 are used to obtain the relationship given in eqn. 7.9
diy
va = xRsix + xLs -j (7.9)

where

and
i, = i.+ (LJL,)it+(Lm/L,)it
The alternative sequence component formulation of eqns. (7.7) likewise leads to
div
va = xRjx + xL\jt (7.10)

where

and

7.2.2 Phase-to-phase and three-phase faults

When the fault involves two or three phases, the voltage between the faulted
phases, say 'a' and 'b', can be found as follows:

dt)
This equation can be reduced to the more succinct form
di
Va~Vb = xRJx + x(La — Lab ) y
-vb = xRaix + x{La-Lab)^- (7.11)
where at

and
{Lb-Lab)ib

When the line is assumed ideally transposed, eqn. 7.10 can be written in terms
of the difference between the currents on each faulted phase, giving
diy
va-vb = xRlix + xLl y (7.12)

where in this case ix = iy = ia — ib.

 Differential equation based techniques 107

The above analysis shows that the behaviour of the transmission line under
fault conditions is governed by a differential equation having the general form of
eqn. 7.13. The measured values of the currents and voltages are taken in the
form of samples and the measurement is usually done simultaneously on all
three phases by using suitable sample-and-hold peripheral equipment as
described previously.
div
v = Rix + L-f (7.13)
at
The values of/? and L in eqn. 7.13 can be calculated by several methods,
described in more detail below.

7.3 Differential equation protection with selected limits

[1-3]
7.3.1 Basic principles
If eqn. 7.13 is integrated once over the time interval t{ to t2 and again over the
period from t3 to /4, the following equations are obtained:

R | \ y / + L(*
* vv22 -*yl
y l)= | vdt
vdt (7.14)

R | \ y / + L (ffyy44--**yy33 ) = rr ^
^ (7.15)
J ti

It will be evident that the values of measured resistance R and inductance L can
be determined by solving these two equations. However, due to the presence of
harmonics, particularly in the current waveforms during the first cycle after the
occurrence of the fault, it has been found that the values of R and L calculated
using this simple approach are very sensitive to low-frequency harmonics. The
accuracy of calculated values of R and L can, however, be greatly improved if
the limits of integration are so chosen as to eliminate (or filter out) the
unwanted harmonics.

7.3.2 Digital harmonic filtering by selected limits

Harmonic components within waveforms can be eliminated if the waveform in
question is integrated over appropriate periods. Let us consider, for example, a
current waveform i(t), which will be assumed periodic in the interval from / = /,
to /, + T. Using the previously described Fourier expansion of eqn. 2.59, i(t) can
be expressed as
a0
COS
i(t) =—+a{ cos ctV + 02 2ftV + a3 cos

+ £, sin a)0t + b2 sin 2a)ot + b3 sin 3<ty+ ' * * (7.16)

108 Differential equation based techniques

If the highest harmonic contained in the waveform i(t) is N, the last equation
would be reduced to

2 (7.17)
where

and

Now let us integrate eqn. 7.17 from ^ = 0 to t2 = a/co0, and let us call this
integration Ix so that:

c
/, = i(t) dt= cQdt^2j m, cos(ma)^+6m)dt
J /,=o
/, = 0 Jo m=\ Jo
This equation shows that the integration of i(t) with respect to time is equal to
the summation of the integration of its individual harmonic components.
Therefore, let us concentrate our attention on the integration of the mth
harmonic component and its multiples. If InX is the integration of the rath
harmonic over the period from tx = Q to t2 = a/(t)0, then
fa/(o0
cn cos(na)0t+dn)dt
J oo
which gives

/ B , = — [sin(raa + 0 B )-sin 0] (7.18)

na)0
Similarly, if we integrate i(t) from t3 = Ji/neo0 to £4= (it In + a)/a)0, and if In2 is the
corresponding integration of the rath harmonic, then
[>/«+a)/a;0

J n

[( B) 8n] (7.19)
na)0
It will be evident from eqns. 7.18 and 7.19 that the sum /„, + /n2 = 0. This is of
course true for all harmonics (n). In essence, this means that any rath harmonic
and its multiples can be filtered out from the waveform i(t), by simply adding
two integrals taken once over the limits /, = 0 to t2 = a/a)0 and t3= (7i/n)/a)0 to
 Differential equation based techniques 109

7.3.3 Graphical interpretation of digital filtering by integration over selected

limits
The filtering of harmonics from waveforms by using integration over selected
limits can be usefully explained graphically. Consider a waveform i(t) and
assume that it consists of only fundamental and third harmonic components,
which can in turn be described by

= /, cos((o0t
When i(t) is integrated from ^ = 0 to t2z=a/a)0 and from J3= (jt/3)/(O0 to
tA= (jr/3 + a)/ft>0 the corresponding integrations of the third harmonic compo-
nents are equal to the areas shown shaded in Figure 7.2. These areas are equal
to each other but have opposite signs and, when added together, they are
cancelled and therefore effectively filtered out from the original waveform i(t).

7.3.4 Filtering of multiple harmonic components

We have seen how it is possible to eliminate, or filter out, nth harmonic
components by simply adding two integrations together. This is because the nth
harmonic contained in the first integration period is cancelled out if the second
integration is performed over limits that are related torathharmonic order. The
same basic approach can be used to eliminate any other harmonic m. This is
achieved by simply adding a third integration over limits related to mth order
harmonic. Therefore, the limits under this condition would be from /5 =
(jt/m)/a)0 to t6 = (7z/m + a)/a)0.
The equation used to eliminate nth and mth harmonics simultaneously from a
waveform i(t) would then be as follows:
Ca/(o0 f{a+nlri)la>Q f{a+n/m)/a>0
i(t)dt+ i{t)dt+ i(t)dt (7.20)
Jo J n/ntOQ J JtlmaiQ

Figure 7.2 Physical interpretation of digital filtering by integration over selected limits
110 Differential equation based techniques

However, the number of integrations can be reduced to only two if the value of
a is chosen to be equal to the angle corresponding to a full cycle of the mth
harmonic order (i.e. a = 27t/m). In this way we can ensure the elimination of the
effect of the mth harmonic by the first integration, as the integration over a full
cycle of a sinusoid is always equal to zero. It is then only required to eliminate
the effect of the nth harmonic, which can be done by addition of a second
integration. Thus, for removing two harmonics of order n and m together with
any multiples thereof, the following equation would be used:
C2n/m(i)Q C{nln + 27i/m)/(x)Q
i{t)dt+ i(t)dt (7.21)
Jo J n/nojQ

By applying the above described principles to eqn. 7.13, it is possible to

calculate R and L so that any number of harmonics are eliminated. For
example, in order to remove the third and fifth harmonics, eqn. 7.13 can be
integrated over
2JT/5\

and again over the interval

pr/3
CD0' \ (o0

The resulting equation, when added, gives

L\ diy+l diA+R\ ixdt+\ ixdt\

LJ 0 J */3tu0 J LJ 0 J JI/3(O0 J

= vdt+l vdt\ (7.22)

LJ 0 J jr/3<y0 J
In general, this principle can be extended to any number of harmonics by
making a sufficient number of integrations.

7.4 Simultaneous differential equation techniques [4, 5]

When these techniques are used, the current and voltage at the relaying point
are related by differential equations. The fundamental assumption used in this
approach is that the transmission line can be represented by either a lumped-
series impedance or a single Pi-section.
7AJ Lumped series impedance based algorithms
This algorithm assumes that the current and voltage waveforms contain a DC
component but are otherwise free from high-frequency oscillations. In other
words, it is assumed that high-frequency oscillations are filtered out from the
original faulted waveforms using low-pass filtering. This is equivalent to saying
 Differential equation based techniques 111

i—"ums^
v(t) 1 v(t)
R L

Figure 7.3 Representation of transmission line using lumped-series circuit parameters

that the filtered voltages and currents are produced by a lumped-series

transmission-line model, such as that shown in Figure 7.3, where y, i are voltage
and current waveforms at the relaying point and R, L are the total resistance
and inductance of the line.
Therefore from the basic principles of circuit theory, v(t) and i(t) are related
by
di
v = Ri + L- (7.23)
at
To determine the values ofR and L using this equation, it is basically required
to have at least two sets of voltage and current samples. Let vh ik be the voltage
and current samples at time tk and let At be the sampling time interval. The
derivative of the current with respect to time (di/dt) can be approximately
determined in sample data form by using eqn. 2.33 to obtain
di i*+i-i*-i
dt 2At (7.24)
Therefore, using sample notations, eqn. 7.23 becomes

(7.25)
2At
Similarly, by using the following sets of samples at tk+l, eqn. 7.23 becomes
+ 2 ~ lk
(7.26)
2At
In matrix form, eqns. 7.25 and 7.26 can be combined to give
h+1 h- I
h R Vk
2A/
(7.27)
*'*+2~~1'*
ik+i L Vk+\
2At \
It is evident that this equation can be written in the following short form:
AP = V (7.28)
where
l
**+l k-\
h 2A/
A=
4+i
2A/ _
112 Differential equation based techniques

By solving eqn. 7.27 or 7.28, the parameters R, L can be expressed in terms of

the current and voltage samples as follows:
"i-4-9 *t ) Vk4- 1 \ ^1-4- 1 It-— I /
(7.29)
«*(^+2- ? '*)- 1 \- h-\)

(7.30)
2 h) h i h-\)
It must be remembered that this algorithm produces a succession of estimates of
R and L as the sampled data is acquired, and these estimates can be compared
for convergence on any post fault values.

7A.2 Single PI section transmission line model based algorithms

These algorithms take a more fundamental approach with regard to the
inclusion of high-frequency oscillations, which can occur during faults. This can
be achieved, at least to a limited degree, by including the effect of capacitance of
the transmission line using a single PI section transmission line model (Figure
7.4).

7A3 Development of the algorithm and basic assumptions

Throughout the derivation of this algorithm, it will be assumed that the
magnitude of any arc resistance Rf is so small that the effect of the far end
capacitance C can be neglected. Therefore, with reference to Figure 7.4, the
circuit equation can be formulated as follows:

(7.31)

The current ic is the capacitive current flowing through the lumped equivalent
capacitance at the relay location, and it is related simply to the voltage at the
relaying point by noting that ic=Cdv/dt. Eqn. 7.31 thus takes a second-order
form:
di do Sv
Ri + L RC---LC-^ (7.32)
dt dt dt2

H I—^MS^-

Figure 7.4 Single PI circuit transmission-line model

 Differential equation based techniques 113

To determine R and L from eqn. 7.32 it is necessary to use at least four sets of
voltage and current samples. Therefore, let us assume that at / = tk we have v = vk
and i = i*. Also, by using central finite differences, the current and voltage
derivatives can be determined from the measured samples using eqns. 2.33 and
2.34 as follows:
di
It 2A/
(7.33)
dv
It 2A/
i
(7.34)
dt2 (A/)2
Substituting current and voltage samples at t = tk and their corresponding
derivatives described by eqns. 7.33 and 7.34, we obtain

Similarly, at instants of time tk+u 4 + 2 and 4+3> we obtain

Rik+l + L RC -LC
2&t 2A< (At)2

Rh+2 -RC — LC
2A/

2A/ 2A/ (AO2

The above equations can be written in the matrix form

R
2A/ (A/)2

2to L
2At (A/)2
l
k+3~~h+\
CR
2A/ 2A* (AO 2

CL
2A/ 2A/ (At)2
(7.35)
In short form, eqn. 7.35 becomes:
A
(7.36)
_C DA LCXJ LV2J
114 Differential equation based techniques

where A, B, C and D are 2 x 2 submatrices of the 4X4 coefficient matrix in eqn.

7.35

Since the vectors X and CX are linearly related, eqn. 7.36 can be reduced to
{A -BBrxC)X= {VI -BDlV2) (7.37)
which can be written in a short form
FX = G (7.38)
The vector X in eqn. 7.38 is readily evaluated by forming the matrices F and G
from sampled values. Again this process can be repeated after each sample set
becomes available, thus producing a succession of estimates of resistance and
inductance between the relaying and fault points.

7.5 References
1 RANJBAR, A.M., and CORY, BJ.: 'Algorithm for distance protection'
Developments in power system protection, IEE Conf. Publ. 125, London, March
1975, pp. 276-283
2 RANJBAR, A.M., and CORY, B J. 'An improved method for the digital protection of
high voltage transmission lines', IEEE Trans. 1975, PAS-94, pp. 544—550
3 GILBERT, J.G., UDREN, E.A., and SACKIN, M.: 'Evaluation of algorithms for
computer relaying', IEEE PES Summer Meeting, 1977 Mexico City, Paper A 77 520-
0
4 SMOLINSKI, W.J.: 'An algorithm for digital impedance calculation using a single PI
section transmission line model, IEEE Trans., 1979, PAS-98, pp. 1546-1551
5 JEYASURAY, B , and SMOLINSKI, W.J.: 'Identification of a best algorithm for
digital distance protection of transmission lines', ibid., 1983, PAS-102, pp. 3358-3359
 Chapter 8
Fundamentals of travelling-wave based
protection

8.1 Introduction
In previous chapters, digital algorithms were derived on the assumption that
the current and voltage waveforms are either sinusoids of a single frequency or
sinusoids consisting of decaying/constant DC, fundamental and/or harmonic
components. Many of these assumptions are valid and lead to an acceptable
performance in applications where the line length is limited. However, the
introduction of long EHV/UHV transmission lines and EHV cables into power
systems can produce difficult problems for protective relays. The principal
causes of these problems are:
(i) shunt leakage current due to the increase in shunt capacitance on long
lines and cables
(ii) relatively low frequency transient currents caused by surge travelling
waves following fault inception
(iii) transient currents, at frequencies as low as a few hundred Hertz, caused by
the interaction between the total inductance and capacitance of the system
(iv) a weakly damped transient DC current caused by high system reactance-
to-resistance ratio.
The foregoing considerations, together with a desire to reduce fault-clearance
times to improve the transient stability of electric power systems, has led to
much interest in the development of so-called travelling-wave protection.
In this Chapter, transmission lines are treated as distributed circuits, in order
to explain how travelling-wave phenomena are propagated and detected. The
basic principles of travelling-wave schemes, the formation of relaying signals,
Bergeron's equation and discriminant functions using single-phase line models
are also explained. These principles are then extended to three-phase lines by
decomposing the line into three equivalent single-phase lines using the modal
decomposition approach.

8.2 The transmission line as a distributed component [1,2]

Travelling-wave methods are usually more suitable for application to long lines
(typically lines with lengths of more than 250 km). Under such circumstances
116 Fundamentals of travelling-wave based protection

dva
R 'a

dvb
ib < 'b dib
w

dvc
k

1 dx

R a dx L a dx
«a ~ di a , di a = di a1 + d i ^ + di a3

di a iV di.

D
1Cadx
Va
G a dx

R
ib »dx ib - di b , di b = di b1 + di b2 + di^

G b dx

ic - di c , dic = di c1 + di c2 + di c

Gcdx C c dx

dx

Figure 8.1 (a) a three-phase transmission line as a distributed circuit

(b) representation of an element dx

the effect of capacitance has to be included in the representation of the line as a

distributed element.
For simplicity, let us ignore losses in the earth return and consider an
elemental length of the three-phase line shown in Figure 8.1. Assume that Rk,
Gk, Lh Ck are the self-resistance, conductance, inductance and capacitance per
 Fundamentals of travelling-wave based protection 117

unit length of phase k (where k = a, b and c) and Lkm, Ckm are the mutual
inductance and capacitance, respectively, per unit length between kth and mth
phases.
If the line is subjected to transient conditions such as faults, the associated
changes in the voltages and currents within an element dx of the line (which is a
distance x from the relay location) at an instant t are related by the following
equations:
For the 'a' phase:

dva dib dic

— —

dia
Cabdx Ccadx —
at at

By considering the line transient at an instant of time / (i.e. dt = O), we can

derive the current and voltage relationships for phase 'a':

dva dib dic

(8.1)

H
Similar equations can be derived for phases 'b' and 'c':
For phase 'b':

dvb dia 8ic

(8.2)

For phase 'c':

(8.3)
118 Fundamentals of travelling-wave based protection

By replacing vah van vba, vbn vca, vch by their equivalent phase voltages (i.e.
v
ab = va~~vb> • • • , etc.), eqns. 8.1-8.3 can be written:
dv
~dx"
(8.4)
di
~d~x=yv
where

»«" i.
v = «= h

and
a a a '
Lca
L
'Jt L b
°m Jt
a a Lb
z— 'lt
a a a
*

a a Cca
a
~dt ~dt
a a a
Gb+Cbb-
It at
a a a

where

Cbb ~ Gab + Cb + Cbc

ca=cca+cbc+cc
By replacing the time operator d/dt in eqns. 8.4 with the transform operator,
the latter can be written as

dv
T
ox (8.5)
di
—
dx
 Fundamentals of travelling-wave based protection 119

where Z and Y are the same as z and y except that the time operator d/dt has
been replaced by the transform operator p. By combining eqns. 8.5 we finally
obtain

(86)
l?=Pv
dH
-2=PH (8.7)

where

PT=YZ
For the purpose of this and the following Chapters, the formulation of eqns. 8.6
and 8.7 is particularly useful for an appreciation of the important special cases
of single-phase lines and three-phase transposed lines.

8.2.1 Travelling waves in assumed lossless single-phase lines

Power lines are of course normally of the three-phase type. However, it is much
simpler to understand travelling-wave concepts and associated methods by first
considering wave propagation in single-phase lines. The equations describing
single-phase lines can be obtained directly from eqn. 8.5 by taking their scalar
version:
dv
- = Zv
ox
(8.8)
di

where v and i are the voltage and line current, respectively, measured at any
point on the line and Z = /? + Lp, Y= G+ Cp. In this case the previously defined
line parameters R, G, L, C represent single-phase line parameters per unit
length.
When eqns. 8.8 are combined, they produce the scalar form of eqns. 8.6 and
8.7, such that

(8.9)

where y is known as the propagation constant, and its value is given by

120 Fundamentals of travelling-wave based protection

By solving eqn. 8.9 as an ordinary differential equation in x and then finding i,

we obtain
v(x)=k{ €~ yx + k2zYX
and

where kx and k2 are constants.

These equations are valid only at some fixed time, i.e. it is assumed that v(x)
and i(x) are functions of the distance x along the line. However, since the line
voltage and current are functions of the time t as well as the distance x, the
equations need to be modified by replacing v(x) and i(x) with v(x, t) and i(x, t),
and the constants kx and k2 by the time functions Fx(t) and F2(t), such that
v(x, /) = e-y*F,(0 + e+y*F2(0 (8.10)

For an assumed lossless line, R = G = 0, which in turn leads to

Vzy=^VZc=- = y and
c
<z [Z
v(x, I) = e-t'T, (1) + c+xp'cF2{t) (8.12)
i(x, t) = (<T*"Fi{t)-e+'*"F2{t))IZ* (8.13)
where
1
C=
VLC
is known as the velocity of propagation and

J
"\c
is known as the surge impedance.
It is possible, with the help of Taylor's theorem, to show that

If this relation is applied to eqns. 8.12 and 8.13 we obtain

v(x,t)=FL--)+F2(l + -) (8.14)

x, t) =
 Fundamentals of travelling-wave based protection 121

Functions Fx{t — x/c) and F2(t + x/c) thus represent travelling waves in the
forward and backward directions of x, respectively, and eqns. 8.14 and 8.15 may
therefore be written in the simple terms

where v+ =Fl(t — x/c), and v~ =F2{t + x/c) are the forward and backward
voltage component, respectively. The values of i+ and i~ are similarly the
forward and backward components of current.
The physical meaning of the function having the form f(t ±x/c), can be well
understood if consideration is given to the behaviour of this function when the
argument is held constant, i.e. let t — x/c = ki for a forward travelling wave and
/ 4- x/c = k2 for a backward travelling wave. Therefore any change in time t, say
At, requires a change in x equal to Ax such that Ax = cAt for the forward
waveform and Ax = — cAt for the backward waveform. Such considerations
show that a forward waveform travels in the positive direction of xy while the
backward waveform travels in the negative direction of x.
The forward and backward components are related to each other, as seen
from eqns. 8.14 through 8.16, by the characteristic (or surge) impedance of the
line as follows:
v+ = Zoi+
- „ ._ (8-17)
V =—ZQI

It will be seen from the above analysis that the surge impedance (Zo) is a real
number for an assumed lossless line and it is evident from eqns. 8.17 that the
current components are simply a replica of their corresponding voltages. They
also show that, while the forward voltage and current waveforms are of the
same sign, the backward voltage and current waveforms are of opposite sign as
illustrated in Figure 8.2.

Coefficient of reflection
Waves travelling over assumed homogeneous lossless lengths of transmission
line continue to propagate at a uniform velocity c and are unchanged in shape.
However, at points of discontinuity, such as open circuits or other line
terminations, part of the incident wave is reflected back along the line and part
is transmitted into and beyond the discontinuity. The wave impinging on the
discontinuity is often called an incident wave and the two waves to which it
gives rise are normally referred to as reflected and transmitted waves.
Consider Figure 8.3, which shows a transmission line whose sending end is
connected to a source and its receiving end is connected to an impedance Zr.
When the switch S is closed, the voltage wave v travels along the line until it
reaches the discontinuous receiving-end terminal where only part of the wave
passes to the terminal impedance Zr (transmitted wave), and the rest is reflected
back. Let us call the first part v* and the second part v~. Therefore, at the
122 Fundamentals of travelling-wave based protection

receiving end the total voltage and current are equal to the sum of the forward
and backward components of voltage and current:
v = v* + v~
(8.18)
,- = ,•+ + ,-
We recall from the previous Section that the voltage wave and its associated
current are interrelated by the surge (or characteristic) impedance of the line:

(8.19)

Combining eqns. 8.18 and 8.19 results in

and solving for v~ we obtain

v =-
z-z0 (8.20)

where

' zr+z0

7777777777777777777777/ 77777777777777777777777

/77/7///////7//77//77T/

Figure 8.2 Propagation of travelling waves on assumed lossless single-phase lines

(a) transmitted voltage and current waveforms
(b) reflected voltage and current waveforms
 Fundamentals of travelling-wave based protection 123

7i
j

Figure 8.3 Successive wave reflection at both ends of a single-phase transmission line
(a) line circuit
(b) wave applied at the source and reflected at the load end

The constant Kr is often called the receiving end coefficient of reflection because
it describes the voltage reflected at a discontinuity in terms of the incident (or
forward wave component).
By combining eqns. 8.19 we similarly obtain
i~ v~ Zr — Z o

or
r =-^+ (8.21)
In a similar manner, the coefficient of reflection at the sending-end can be
derived as:
Z
" '~Z° (8.22)

where Zs is the sending end source impedance.

8,2.2 Three-phase transposed lines
When a three-phase line is perfectly transposed, the parameters are identical for
each phase. In common with untransposed lines, such lines can be decomposed
into three modal equivalent circuits.
Modal analysis of three-phase transmission lines [3, 4, 5]
With reference to eqns. 8.6 and 8.7, it will be seen that the elements of F = ZF
are independent of x. It follows that eqns. 8.6 and 8.7 can be treated as linear
124 Fundamentals of travelling-wave based protection

differential equations with respect to voltages and currents. On the other hand
it is well known in power system analysis that phase quantities can be
decomposed into equivalent symmetrical components such that

where Vp, Ip are the phase voltage and phase current vectors, V^ Is are the
sequence voltage and sequence current vectors and T is the symmetrical
component transformation matrix defined below:
1
vb , v = and T =
1Z_12O°
are tne
K> Vfo K voltages of phases 'a', ' b ' and V , and F o , F,, F2 are the zero,
positive and negative sequence voltages.
It should be noted that, in accordance with the theory of symmetrical
components, phasor values are employed throughout in the above analysis.
It is possible likewise to transform the instantaneous (or time domain) phase
quantities into what are called modal components such that
vp(t)=Svm(t)
(8.23)
iP(t) = Q.L(t)
In this case, the voltage and current vectors vm(t), im(t) are modal component
vectors defined by
f(0)
(I) (t)
(0 I (<)
w. I w
It should be noted that the superscripts (0), (1), and (2) refer to the zero, first
and second modes. The matrices S and Q are the voltage and current
transformation matrices, which are used to transform the phase quantities into
modal components.
The elements of matrices S and Q can be determined using matrix function
theory and the properties of eigenvalues and eigenvectors. Details of the
analyses involved are beyond the scope of this book, but a brief outline of the
method is useful.
By substituting eqns. 8.23 into eqns. 8.6 and 8.7, we obtain

(8.24)
dx2

— = Q-ipTQim (3 25)

It is necessary to choose the matrices S and Q such that S~lPS and Q~lPTQ
become diagonal. This involves finding three scalar quantities (the eigenvalues)
 Fundamentals of travelling-wave based protection 125

and a column matrix X (the eigenvectors) for eqns. 8.6 and 8.7. For example,
the eigenvalues and eigenvectors for eqn. 8.6 should satisfy the following
equation:

{P-U)X=0 (8.26)

where U is the unit matrix.

For perfectly transposed lines, the matrices S and Q are equal to each other
for some transformations, and in protection applications, three modal transfor-
mation matrices have been widely used. These are:
1. The so called Wedepohl transformation

1 1 1
1 0 -2 (8.27)
1 -1 1^

2. The Karrenbauer transformation

ioni
"l 1 r
1 o i (8.28)
1 1 -2_

3. The Clark transformation

1 1 0
1 -1/2 -3/2 (8.29)
1 -1/2 -3/2

Modal surge impedance

Because the transformation matrices S and Q are real for a transposed line, they
can be applied in both the frequency and time domains. Thus, by substituting
the foregoing transformations (S = Q) into eqns. (8.24) and (8.25) a set of
decoupled propagation characteristics can be obtained:

IF
(8.30)
IF

where y(i) is the modal propagation constant of the kth mode. The zero mode is
126 Fundamentals of travelling-wave based protection

also termed the earth mode, while the second and third modes are termed
'aerial modes'.
These equations show that wave propagation in a three-phase line can be
considered in terms of three separate and independent components, each with
its own propagation constant y{k) and associated surge impedance Z$\
Assuming complete transposition, the surge impedance associated with each
mode can be shown to be:

40> = ( J J ( f m ) J0
2>
4 " = 4 = V(z,-zm)/(Y,+ Ym) = VzJY,
where Z0(k) is the A>mode surge impedance, Zs, Ys are the average sum of all
conductor self impedances and admittances at any frequency, Zm, Ym are the
average sum of all conductor mutual impedances and admittances at any
frequency, Zo, Yo are the zero-phase sequence line impedance and admittance
and Z l5 Yl are the positive-phase sequence line impedance and admittance.
The modal propagation constants are similarly given in terms of phase
sequence impedances and admittances by

(8.32)

(8.33)

8.3 Superimposed quantities and their properties

When a power system is subjected to a fault condition, the total voltage and
current at any point in the system can be considered as consisting of two
components: one is due to the sinusoidal steady-state condition, the other due to
the application of the fault. The latter component is often called the superim-
posed quantity and is simply equal to the change in the current and/or voltage
due to a disturbance. The idea of superimposed quantities is best explained by
reference to Figure 8.4. Consider point R, where the voltage vRf(t) and current
iRj(t) are the actual quantities present (see Figure 8.4a). These quantities can be
split into the sinusoidal prefault quantities vRS(t) and iRs(t) shown in Figure 8.4£,
and the fault superimposed quantities AvR(t) and AiR(t) shown in Figure 8.4c.
Thus

It should be noted that the sum of the steady-state and superimposed quantities
is equal to the total variation at ail points in the network. The superimposed
 Fundamentals of travelling-wave based protection 12 7

network is therefore a network with all source voltages set at zero (short
circuited). The superimposed fault point voltage is therefore zero for ail time up
to the time of the fault disturbance. The voltage of the fictitious superimposed
source on the faulted phase(s) is thus equal in magnitude and opposite in sign to
the prefault voltage at the fault point (i.e. vF(t)).
The superimposed quantities of voltages and currents possess unique
properties in relation to the location of the fault, which will be explained in the
following subsections. These properties form the basis of practical travelling-
wave based protection schemes.

83.1 Polarity of superimposed quantities versus fault location [6, 7]

We shall show in this Section that the relative polarities of the superimposed
voltage and current waveforms (or the direction of the motion of their travelling
waves) at the two ends of the line under consideration depend on the fault
location. Consider Figure 8.5, from which it can be seen that the fault produces
superimposed quantities at the relaying points R and S. If an internal fault is
assumed to occur during the positive half cycle of the prefault voltage
waveform, then the signs of the superimposed voltage and current would be
( - ) and ( + ), respectively, at both ends of the line. (Note that reference

o i > ; <i o
S

(t) r

Figure 8A Superimposed voltage and current generated by a fault

(a) actual system under a fault condition
(b) the steady-state (prefault) network
(c) the superimposed network
128 Fundamentals of travelling-wave based protection

o-
9 is incidence angle

o-
polarity of
superimposed
type of fault corresponding superimposed
network polarity components
AiR AvR Ais Avs

pos.

internal fault
neg.

pos.

external fault
beyond bus R
neg.

pos.
external fault
beyond bus S
neg.

Figure 8.5 Dependency of polarities of superimposed voltage and current quantities on fault
location

directions are those shown in Figure 8.4). However, if the fault occurs during
the negative half cycle of the voltage waveform, the polarities of the superim-
posed voltage and current at the R and S ends would be ( + ) and (—)
respectively. Repeating the analysis for faults at locations external to the line
R—S reveals that the polarities of the superimposed voltage and current would
be similar to each other at one end but are different from each other at the other
end. Figure 8.5 explains in detail all possible cases.
 Fundamentals of travelling-wave based protection 129

8.3.2 Interrelation between the superimposed voltage and current quantities versus
fault location [8, 9]

Consider Figure 8.6, which shows a line R-S interconnecting two systems R
and S. The following analysis and associated conclusions are general, but to
simplify the discussion we shall assume that

(i) the line is lossless, and therefore the surge impedance of the line (Zo) and
the wave propagation velocity c are given by

(8.35)

(ii) the surge impedances looking into systems S and R are equal to Zos and Z0R,
respectively, and are such that

— Z0S—Z0/3 (8.36)

These terminating conditions are exactly analogous to a situation in which each

substation terminates four lines of equal surge impedance Zo. In the light of
these assumptions, which are used in producing Figures 8.7 and 8.8, we shall
examine the relationship between the superimposed voltages and currents for
faults in the reverse and forward directions as seen by a relay located at end R.

Reverse faults
Let us assume that a solid fault occurs immediately behind the relay location at
R. Now because the fault point coincides with the relaying point, the resulting
superimposed voltage at R, AvR(t), is equal and opposite to the steady-state
prefault voltage waveform Vp(t) at end R, i.e. AvR(t) = —Vp(t). As discussed in
Section 8.2.1, this causes a travelling wave to propagate towards end S at a
velocity c defined by eqn. 8.35. The associated superimposed current waveform
AiR(t) also propagates at a velocity c and the superimposed voltage and current

forward reverse
direction direction

777T

Figure 8.6 Line interconnecting two systems R and S

130 Fundamentals of travelling-wave based protection

ZQS = ZQ/3
AiR(t)

+
)Av F (t) = AvR(t)

distance
fromR

2l/c time after fault

Figure 8.7 Initial propagation of superimposed quantities following a reverse fault

immediately behind the relay at R
(a) superimposed fault network
(b) lattice diagram
(c) time of variation of superimposed quantities consistent with fault at the
positive-to-negative zero-crossing of the prefault voltage
total superimposed voltage at R
AvR(t) = AvF(t) = h(t) Vm sin coQt
initial superimposed current AvF(t)/Z0
total reflected superimposed current in line at R at time 21/c,
AA21/)/^
total superimposed current in line at R
AiR(t) = AvF(t)/Zo + AvF(t-2l/c)/ZQ

waveform AvR(t) and AiR(t) are interrelated according to eqn. 8.17. In the
period of time from fault inception up to twice the wave transit time from end R
to S {211 c, / = length of line), AvR(t) = Av^(t), AiR(t) = A$(t) and AvR(t) =
 Fundamentals of travelling-wave based protection 131

distance
from R

time after fault

Xp/C

Figure 8.8 Initial propagation of superimposed quantities following a fault in the forward
direction of the relay at R
(a) superimposed fault network
(b) lattice diagram
(c) time variation of superimposed quantities consistent with fault at
negative peak of prefault voltage
superimposed fault voltage vF(t)
superimposed voltage at R
vR(t) = AvF(t-xF/c)/2
superimposed current at R
iR(t) = -3AvF(t-xF/c)/2Zo
132 Fundamentals of travelling-wave based protection

AiR(t)=O; it therefore follows that, during this period, the superimposed

voltage and current at the relaying point R are related by
AvR(t) = Z0AiR(t) \
f (8.37)
or A(t)Z£i(t) 0)
Because S is a point of discontinuity, part of the incident superimposed voltage
and current waveforms are reflected back towards the relaying point at R,
according to eqns. 8.20 and 8.21, such that

s ( ) r H )
(8.38)
Ag(t) = -KrAiHt)
where
Kr=(Z0S-ZQ)/(Z0S+Z0)
and Avs{t), Ais(t) are the transmitted parts of superimposed voltage and
current waveforms, respectively, at end S, and AvJ(t), Ais(t) are the reflected
parts of the superimposed voltage and current waveforms, respectively, at end
S.
It will be evident that it takes a time of 2 He for the waveform reflected from
end S to arrive back at R after the occurrence of the fault. Figure 8.7 thus
illustrates the behaviour of the system shown in Figure 8.6 due to a fault that
occurs at the positive-to-negative zero crossing of prefault voltage. Figure 8.7(b)
shows the lattice diagram of the superimposed voltages and currents at any
point on the line, and Figure 8.1 (c) shows the resulting superimposed compo-
nents AvR(t), AiR(t) at the relaying point R. It is particularly important to note
that if it is assumed that the fault occurs at time / = 0, then
(i) AvR(t) = AiR(t) = 0 for t < 0
(ii) Although the waveforms of Figure 8.7 (c) are consistent with a fault at zero
voltage point on wave, the relationship of eqn. 8.37 also holds for all time
from fault inception up to twice the wave transit time, i.e. 21/c for any
arbitrary fault point superimposed voltage waveform AvF(t).
(iii) The relationship described in eqn. 8.37 also holds for any value of
terminating surge impedances (ZQRi Zos) and line attenuation. For exam-
ple, if the source surge impedance at end S were exactly equal to that of the
line, there would be no reflection from S, but eqn. 8.48 still holds for all
times
For faults within system R, there is a time delay between fault inception and the
arrival of the superimposed components at the measuring point. Nevertheless,
eqn. 8.37 holds for all times t = 2l/c after the arrival of the superimposed
quantities at R.
Forward faults
Now let us consider the case where faults occur in the forward direction with
respect to the relay at end R. For example, consider a fault at a point F which is
 Fundamentals of travelling-wave based protection 133

at a distance xF from R. The resulting superimposed voltage at xFy AvF(t), is

equal and opposite to the prefault voltage at that point. Again in accordance
with Section 8.2.1, the application of the superimposed voltage AvF(t) causes the
voltage travelling wave and its associated current wave Avp(t)/Z0 to propagate
at a velocity c towards the point R. The superimposed voltage and superim-
posed current waveforms incident at R arrive at a time xF/c and are given by

AvR{t) = A^(/) = AvF(t-xF/c)

AiR(t) = - Ai J (/) = - &vF(t - xF/c) /Z o

Now the reflected components of these voltage and current waveforms for an
arbitrary terminating surge impedance Z0R can be found from the correspond-
ing incident waves and the reflection coefficient:

= AvF(t-xF/c)(ZQR-Z0)/(Z0R + Z0) (8.39)

= -AvF(t-xF/c)(Z0R~~Z0)/[(Z0R + Z0)Z0] (8.40)

But the total variation of superimposed voltage and current at, and immediately
after, the time when the waves reach end R (t = xF/c) is given by the sum of
transmitted and reflected waves:

= 2ZORAvF(t-xF/c)/(ZOR + Z0) (8.41)

= -2AvF(t-xF/c)/(Z0R + Z0) (8.42)

Therefore, by finding the difference between eqns. 8.41 and 8.42 and their sum,
we obtain

AvR(t) - Z0AiR(t) = 2AvF(t - xFlc)

and (8.43)
A{t) + ZAi{t) = 2AvF{t-xFlc) (Z 0 / ? ~Z 0 )/(Z 0 / ? +Z 0 )

Eqns. 8.43 show that, unlike the previously considered reverse fault, the
difference AvR(t) — Z0AiR(t) becomes immediately finite on arrival of the
superimposed waves at xFic after fault inception. This conclusion holds for
faults anywhere on the line or within system S.
Figure 8.8 illustrates the behaviour of the superimposed quantities following
a fault on the line R-S of the system of Figure 8.6. The fault occurs at the
negative peak of the prefault voltage. It is evident from the lattice diagram and
134 Fundamentals of travelling-wave based protection

the waveforms of Figures 8.8(£) and (c), that, initially, the travelling waves of
superimposed voltage and current are of opposite polarity.
However, in case of a fault at the end S, there is a time delay between fault
inception and the arrival of the superimposed quantities at R. Although the
waveforms shown in Figure 8.8(c) are consistent with a fault at peak prefault
voltage, the conclusions drawn equally apply for any other arbitrary superim-
posed fault point voltage waveform.

Relaying signals
The difference in the behaviour of the superimposed quantities according to
whether the fault is in the reverse or the forward direction can be used to
determine the direction of the fault with respect to the measuring point. This is
done by the formation of two signals. In terms of primary system values, the
signals used at R are
SlR=AvR(t)~R0MR(t)
(8.44)
S2R=AvR(t)+R0AiR(t)
The corresponding signals used at S which have the reference directions defined
in Figure 8.8(0), are given as
S I S =A^(0-/?oAf 5 (0
(8.45)
S2S= Avs(t) + R0Ais(t)
where Ro is a surge replica resistance whose value is so arranged as to match
closely the line surge impedance Zo.
By examining the signals at end R for reverse faults, with the assumption of
lossless line and perfect matching of the replica resistance Ro, i.e. by substituting
eqn. 8.37 into eqns. 8.44, we obtain
SlR=0
(8.46)
S2R=2AvR(t)
Figure S.9(a) shows how the signals vary following the reverse fault condition
considered previously. It is important to note that S2R becomes finite before SlR.
Figure S.9(a) also shows the sequence in which the signals SlRand S2R rise above
a given threshold value \VS\, (S2R followed by SlR), at times Tx< T2.
Similarly, the signals SlR and S2R for the forward fault described previously
can be found by substituting eqn. 8.43 into eqn. 8.44. By assuming perfect
matching of the replica resistance Ro to the surge impedance Zo, these signals
would be given as
SIR = 2AvF(l-xF/c)
(8.47)
S2R = 2AvF(t-xF/c) (Z0R-Z0)/(Z0R+Z0)
Figure 8.9(6) illustrates the behaviour of the signals SIR, S2R due to the forward
fault condition on the system of Figure 8.8. It can be seen that the setting levels
Vs and — Vs are exceeded almost simultaneously (TlssT2) owing to the very
 Fundamentals of travelling-wave based protection 135

\
1

voltage
Vs Vs
/T, xp/c = Ti = T 2
2I/C

" T2 time after fault time after fault

V
-VS
-V- -Vs

j • • • " /

Figure 8.9 Variation of relaying signals at R following faults on the system of Figure 8.6
(a) reverse fault at R
(b) fault on line R-S
R0MR(t)
SlR = AvR{t) - R0MR(t)
S2R = AvR(t) + R0MR(t)

rapid change in the superimposed components associated with the assumed

fault at peak prefault voltage.
When a three-phase line is considered, the phase variations of currents and
voltages are decomposed into modal variations, and the relaying signals can
therefore be expressed in terms of aerial-mode surge impedances Z^!) and 7§\
and the earth-mode surge impedance Z^o) (see Section 8.2.2). Assuming a
lossless three-phase line with complete transposition and perfect matching of
the replica surge impedances RQ\ k = 0, 1 and 2 to the corresponding surge
impedance Z$\ k = 0, 1 and 2, the modal relaying signals at the end R of the line
are given by

= 0, 1, 2 (8.48)
136 Fundamentals of travelling-wave based protection

8.3.3 Behaviour of relaying signals at the relay and fault locations [10]
The way relaying signals S, and S2 behave at the relay and fault points of
discontinuity can be used to estimate the distance of the fault to a relay.
Consider a forward fault on the line of the system shown in Figure 8.10. The
superimposed voltage at the fault point F is AvF(t) and its associated
superimposed current AvF(t)/Z0 propagates towards end R.
Thus, if the relaying signal SlR (see eqn. 8.44) is formed from the above
voltage and current before and after they are reflected at the relay busbar we
would find its value does not change, and it remains equal to
SlR=2AvF(t-xF/c) (8.49)
The reflected voltage and current waves then propagate back towards the fault,
where they are reflected back again towards end R. Assume the most common
case, where the fault resistance Rj is much smaller than the surge impedance of
the line. Under this condition the coefficient of reflection at the fault point
Kf=(Rf—Z0)/(Rf+Z0)9 would be approximately equal to unity. Therefore,
according to eqns. 8.20 and 8.21, the voltage and current waveforms will be
reflected at the fault with almost the same magnitudes. However, the reflected
voltage would have the same polarity as the incident voltage waveform, while
the reflected current would have opposite polarity. If SlR is recalculated from the
reflected waves, its value would be zero. The important conclusion from this
analysis is that the relaying signal SlR that leaves the relay point toward the fault
would be reflected back to the relay, with its shape and magnitude changed at
the fault point.
A similar analysis can be carried out with regard to S2R, which reveals that it
is changed at the relaying point, in exactly the same manner as the signal SlR. It
follows that, if the time between like changes in SlR and S2R is equal to r, the
distance to the fault xF would be
T/2
xF = (8.50)
c
Figure 8.10 shows the response of a single-phase lossless line to a 1 p.u. voltage
step function applied at the fault point, for values of fault resistance Rj of zero
and of 25% of the characteristic impedance of the line. The coefficients of
reflection are calculated on the basis that inductive sources are connected at
buses 1 and 2, having short-circuit capacities of 35 000 MVA and 10 000 MVA,
respectively.

8.3.4 Superimposed component elliptical trajectories [11]

For the following discussion, let us consider the system shown in Figure 8.11. If
we assume the prefault voltage at the fault point F is vF(t)= V sin(a)0t + 0),
where V and 6 are the peak voltage and the inception angle, respectively, then
the superimposed voltage and current waveforms AvR(t) and AiR(t) at end R can
be calculated as explained below.
 Fundamentals of travelling-wave based protection 137

0.0 0.5 1.0 1.5 2.0 2.5 time, ms

fault resistance = 0 ohmns

AvR(t)
RoAiR(t)

RoAiR(t)-AvR(t)

S2R=RoAi R (t) + Av R (t)

fault resistance = RQ/4

S2R

stored S2R

Figure 8.10 Variation of relaying signal SlR and S2R at the relay and the fault points
(a) system under fault condition with the corresponding lattice diagram
(b) voltage and current waveforms due to a zero resistance fault and the
corresponding signals .S^ and S2R with zero fault resistance
(c) the signals SlR and S2R due to the same fault but with fault resistance
equal to 25% of surge impedance
138 Fundamentals of travelling-wave based protection

L
Ls R f c S

= Vsin(coot

Ls R Lf s
—>—

AVR f ) | -AVF(t) = -Vsintcoot + 6)

Figure 8.11 Representation of single-phase system used to explain the elliptical trajectory
based technique
(a) steady-state network
(b) superimposed network

With reference to Figure 8A\(b), the superimposed voltage — Vsin(a)ot+O)

causes a superimposed current A^(J) to flow according to the following
differential equations:

dAiR(t)

(8.51)

By solving eqn. 8.51, we obtain

MR(t)=- (cos(fty+ 6) — cos 6)

o>0{Ls

or
AiR(t) = A/(cos(tty + 0) -cosO) (8.52)

where

A/=
a)0(Ls
 Fundamentals of travelling-wave based protection 139

Again with reference to Figure 8.11(£), the superimposed voltage Az^(/) can
be determined from the superimposed current AiR(t):
dMR(t)
AvR(t)~-Ls~j-~ (8.53)

By substituting eqn. 8.52 into eqn. 8.53 we obtain

AvR(t) = a)0LsAI sin(oV+ 0)
or
AvR(t) = AVsin((Dot+0) (8.54)
where
xy
As- f

and XS = O)QLS is the reactance of the source, and Xj= (OoLfis the reactance of the
line to the fault.
Multiplying eqn. 8.52 by a mimic resistance R and combining the resulting
equation with eqn. 8.54 results in
Av2R(t) (RAiR(t)-RAIcos0)2
+
"(Alf (RAW -1 (8 55)
*
By examining eqns. 8.52 to eqn. 8.55, the following conclusions can be drawn:
(i) Eqn. 8.52 shows that the superimposed current component AiR(t) contains
a DC offset equal to A/ cos 0, which depends on the fault inception angle
e.
(ii) Inclusion of resistance in the faulted loop will, of course, cause a decay of
the DC offset. On the other hand, such resistance will slightly reduce the
magnitudes of A V and A/ and introduce an additional phase shift between
the fundamental frequency components of the superimposed quantities.
(iii) Eqn. 8.55 represents an ellipse in the superimposed plane (i.e. the
RAiR(t)-AvR(t) plane), with its centre being shifted in the direction of
RAiR(t) by a constant value of RAI cos 0. This ellipse is known as the fault
trajectory.
Figure 8.12(a) shows the steady-state voltage vSR(t), the superimposed voltage
AvR and the superimposed current AiR due to the fault under consideration.
Figure SA2(b) shows the fault trajectory. It can be seen that immediately after
the fault, AvR will have a finite value, while AiR (under the assumption of a
lossless line) changes gradually from zero. Therefore, the trajectory initially
jumps from the origin of the plane to a point somewhere along the AvR axis. It
will then proceed along the elliptical path of Figure 8.12(^). According to eqn.
8.54, the magnitude of initial jump AvR is a maximum when the fault inception
angle 0 is equal to ± n7t/2, i.e. the prefault voltage is maximum, while AvR is
zero for a zero fault inception angle.
140 Fundamentals of travelling-wave based protection

VSR(t)

AV RAi

RAI
forward fault

RAI COS 6

reverse fault

Figure 8.12 (a) Behaviour of voltage and current waveforms under fault
conditions
(b) corresponding fault trajectory

8.4 Bergeron's equations [12, 13]

8.4.1 Single-phase lines
To simplify the analysis once more, a lossless single-phase line will be
considered so that, by combining eqns. 8.14 and 8.15, we obtain

«"(*, t) + ^ v(x, t) = 2FL--\ /Z0 (8.56)

 Fundamentals of travelling-wave based protection 141

iR(t) «s(t)
—*—O

vR(t) vs(t)

Figure 8.13 Single-phase line used to explain Bergeron's equations

If we consider the transmission line shown in Figure 8.13, eqn. 8.56 can be
applied to the R and S ends, where x = 0 and x — l, respectively. The following
equations are then obtained:

i(O,t)+jv(O,t) = '. (8.57)

1
(8.58)

By renaming i(0, /), v(0, t) as iR{t) and vR(t), i(l, t) and v(l, t) as -is(t) and vs(t),
substituting / by (t — l/c) in eqn. 8.57, and combining the resulting equation
with eqn. 8.58, we obtain Bergeron's equation:
1 1
iR(t - r) + — vR(t - r) = - is(t) + — vs(t) (8.59)
Zo Ao

where r = l/c is the surge transit time.

The right-hand side of eqn. 8.59 is twice the forward component of the
current originating from end R of the line at time (t — i) (see eqn. 8.57).
Similarly, the left-hand side of the equation equals twice the same forward
current component originating from end R and arriving at S at time /, i.e. after a
time delay of r where r = l/c, which is the travelling time of the forward current
component over a line of length /.
Eqn. 8.59 therefore means simply that the forward current component
originating from one end of a healthy line arrives at the other end after a time
delay equal to the surge time of the line.
A further 'Bergeron's equation' can be obtained by eliminating F2(t + x/c)
from eqns. 8.14 and 8.15, and this in turn results in
1 1
— vR{t) = is(t - r) + — vs(t - r) (8.60)

The physical interpretation of this equation is that a backward component

originating from one end of a healthy line arrives at the other end after a time
delay equal to the line surge time.
142 Fundamentals of travelling-wave based protection

8.4.2 Three-phase lines

To find Bergeron's equations for a three-phase line, the phase variations of

currents and voltages are first decomposed into modal components. Each modal
equation is then treated in exactly the same manner as for a single-phase line. If
the line is assumed to be lossless and ideally transposed, eqns. 8.5 can be
reduced in terms of line capacitances and inductances as follows:
dv
ox
(8.61)
di

When these equations are decomposed into their modal components as

explained in Section 8.2.2, we obtain

=Lmim
dx
(8.62)

where
L (0) 0 0 c0) 0 0
0 £ (1)
0 cm= 0 C (l) 0
(2) 2)
0 0 z 0 0 c<
In fact, eqns. 8.62 represent independent sets of modal equations. Therefore,
the derivation procedure outlined previously for the single-phase line is
applicable to each of these modal equations. Thus, for a three-phase line,
Bergeron's equation can be derived from eqns. 8.59 and 8.60 as follows:

(8.63)
1

where r{k) is the k mode surge travellin is the k mode surge

impedance, r w = /V'L{k)&k) and Zf = /
 Fundamentals of travelling-wave based protection 143

8.5 Discriminant functions [5, 14, 15]

It is possible to differentiate between internal and external (or forward and

reverse) faults with respect to a power line terminal by employing suitable
discriminative functions. The basis of this technique is given in this section.

8.5.1 Single-phase lines

Let us again consider a faulted lossless single-phase line (Figure 8.14(a)). As
established previously, the fault can be simulated as a sudden application of a
superimposed voltage AvF(t) at the fault point F, where AvF(t) is equal and
opposite to the prefault steady-state voltage at point F. This causes a
superimposed voltage AvR(t) and current AiR(t) to appear at end R.
By applying Bergeron's eqn. 8.59 to the section R—F of the line, and
substituting AvF, AiF, AvR, AiR and (/ + r) for vR, iR, vs, is and t, we obtain
AvR(t + r) - Z0AiR(t + r) = AvF(t) + Z0AiF(t) (8.64)
Now let us assume the prefault voltage at the fault point F, immediately before
the occurrence of the fault, is vF(t), given by
vF(t) = Vsin(a)ot + 0)
Therefore
A MO = - Vs sin(a>ot-¥0) (8.65)
where V and 6 are as defined previously.
Since the backward or reflected travelling wave b shown in Fig 8.14(rf) is
initially zero, the current change AiF(t) at the fault point during the period
between the arrival of the wave at end R and its return back to the fault point F
will be given as follows (see eqns. 8.14 and 8.15):

A)

(8.66)
The forward wave/, i.e. Av*(t) and/or Ai*(t)y which propagates towards end
R, and the backward (or reflected) wave b, which is reflected back towards the
fault point F, are shown in Figure 8.14(rf).
By substituting AvF(t) and AiF(t)y given by eqns. 8.65 and 8.66, respectively,
into eqn. 8.64 we obtain
AvR(t-¥T)-Z0AiR(t + T) = -2Vsm(<t)Qt + d) r^t^2t (8.67)
It can be seen that eqn. 8.67, which describes the wave characteristics as seen at
end R, depends on the fault inception angle 6. This dependency can be removed
144 Fundamentals of travelling-wave based protection

by taking the derivative with respect to time of the wave characteristic eqn.
8.67, which in turn gives

- (AvR(t) - Z0MR(t)) = - 2(o0V cos(ft>0/+ 0') (8.68)

where 6' = 0 —a>ftt.

IFS

•O
v F (t)

"Ft
-<—
F

o
AiR(t) AiF(t) F
->

Av R (t) Av F (t)

2T

Figure 8.14 Single-phase system used to explain the forward discriminant function
(a) system under steady-state conditions
(b) system under a forward fault condition with respect to a relay at end R
(c) superimposed network
(d) the corresponding forward and backward waveforms,/and b respecti-
vely
 Fundamentals of travelling-wave based protection 145

o o
iRS(t)

VRS(t)

-o
V f VRf(t)

Av F (t)
AVR
= -V F (t)

Figure 8.15 Single-phase system used to explain the backward discriminant function
(a) steady-state system
(b) system under a reverse (or backward) fault condition with respect to a
relay at end R
(c) the corresponding superimposed network
(d) the corresponding forward and backward travelling waveforms,/and
b respectively

This equation can be combined with eqn. 8.67 to give eqn. 8.69, which
describes the wave characteristic as seen by a relay located at end R due to a
fault on the line:
1 Yd V
]2
DF= (AvR(t) -Z 0 A^(/)) 2 + — - (AvR(t)-Z0MR(t))
ujQ^at j

= 4 ^ = 8 K22RRMS (8.69)
146 Fundamentals of travelling-wave based protection

where FRMS is the RMS value of the supply voltage and DF is known as the
'forward wave discriminant function'. Repeating the analysis for faults that
occur behind the relay located at end R, as shown in Figure 8.15, results in
another function, which is often called the 'backward wave discriminant
function' DB

{t))2 + \ \ j (AvR(t) + Z0AiR(t))Y

°L J
= 8 F| M S forr,^^r l + 2r (8.70)

8.5,2 Three-phase lines

The discriminant function for three-phase lines can similarly be found in terms
of the superimposed modal voltage and currents at the relaying points:

* = 0, 1,2 (8.71)

for the modal (k) forward discriminant function, and

k = 0, 1,2 (8.72)

for the modal (k) backward discriminant function.

ZQ^ is the mode k surge impedance, and Av$ and Aijf are the modal k
superimposed voltage and current at the relay location end R.
In practice, algorithms based on this technique can be very sensitive to noise
because of the need to use time derivatives. This problem can be ameliorated in
some applications by suitable low-pass pre-filtering of the measurements.

8.6 References
1 BEWLEY, L.V.: 'Travelling waves on transmission systems (2nd edition)' (John
Wiley, 1951)
2 BICKFORD, J.P., MULLINEUX, N., and REED, J.R.: 'Computation of power
system transients' (IEE Monograph Ser. 18, Peter Peregrinus, 1976)
3 WEDEPOHL, L.M.: 'Application of matrix methods to the solution of travelling-
wave phenomena in polyphase systems', Proc. IEE, 1963, 110, pp. 2200—2212
4 DOMMEL, H.W.: 'Digital computer solution of electromagnetic transients in single
and multiphase networks', IEEE Trans. 1969, PAS-88, pp. 388-396
5 DOMMEL, H.W., and MICHELS, J.M.: 'High speed relaying using travelling
wave transient analysis', PES Winter Meeting, 1978, Paper A78 214-9
6 CHAMIA, M., and LIBERMAN, S.: 'Ultra high speed relay for EHV/UHV
transmission lines—development, design and application', IEEE Trans. 1978,
PAS-97, pp. 2104-2116
7 YEE, M.T., and ESZTERGALYOS, J.: 'Ultra high speed relay for EHV/UHV
transmission lines- installation-staged fault tests and operational experience', ibid,
pp. 1814-1825
 Fundamentals of travelling-wave based protection 147

8 JOHNS, A.T.: 'New ultra-high-speed directional comparison technique for the

protection of E.H.V. transmission lines', IEE Proc. C, 1980, 127, pp. 228-239
9 JOHNS, A.T., MARTIN, M.A., BARKER, A., WALKER, E.P., and CROSSLEY,
P.A.: 'A new approach to EHV directional comparison protection using digital signal
processing techniques', IEEE Trans. 1986, PWRD-1, pp. 24-34
10 CROSSLEY, P.A., and McLAREN, P.G.: 'Distance protection based on travelling
waves', IEEE Trans. 1983, PAS-102, pp. 2971-2983
11 VITINS, M.: 'A fundamental concept for high speed relaying', IEEE Trans. 1981,
PAS-100, pp. 163-173
12 TAKAGI, T. et al.: 'Fault protection based on travelling wave theory, Part I: theory',
PES Summer Meeting, Mexico, 1977, Paper A77 750-753
13 TAKAGI, T., et al.: 'Fault protection based on travelling-wave theory, Part II:
sensitivity analysis and laboratory test', PES Winter Meeting, 1978, Paper A78 220-
226
14 MANSOUR, M.M., and SWIFT, G.W.: 'A multi-microprocessor based travelling
wave relay—theory and realization', IEEE Trans. 1986, PWRD-1, pp. 273-279
15 MANSOUR, M.M., and SWIFT, G.W.: 'Design and testing of a multi-
microprocessor travelling wave relay', ibid, pp. 74-82
 Chapter 9
Travelling-wave protective schemes

9.1 Introduction

The previous Chapter laid down the foundation of travelling-wave protective

methods. This Chapter is designed to extend knowledge about travelling-wave
techniques by presenting the underlying principles of some specific implemen-
tations. Information on further development of specific products is available
from manufacturers' literature.

9.2 Bergeron's-equation based scheme [1,2]

9.2.1 Principles of internal fault detection

The principle of detecting an internal fault on a transmission line can best be
explained using a single-phase lossless line model, such as that shown in Figure
9.1 (a), whose voltages and currents at the two terminals of the line are related
by Bergeron's eqns. 8.59 and 8.60; these are rewritten below:

1 1
**(' " T) + — VR(t - T) = - is(t) + - vs(t) (9.1)
A) A)

1 1
- vR{t) = is(t - r) + - vs{t - r) (9.2)
A ^

As explained in Chapter 8, eqn. 9.1 implies that a travelling wave leaving end R
will arrive at end S after a time delay of r, while eqn. 9.2 implies that a
travelling wave leaving S will arrive at R after a similar time delay. For faults
external to the line, eqns. 9.1 and 9.2 are satisfied. However, when a fault occurs
 Travelling-wave protective schemes 149

Figure 9.1 Single-phase line

(a) with no fault
(b) with internal fault

on the line, they are not satisfied, because a discontinuity occurs between each
end. Thus, the equations can be written in the form of eqn. 9.3, in which
£j(/) = o, e2(t) = 0 for an external fault and
£,(/)=£(), ^ W ^ O f° r an
internal fault,
where
1
I & iR(t - r) + is(t) + — [vR(t -T)-v:
-A (9.3)
1
& is(t - T) +«« [vs(t - r) -
150 Travelling-wave protective schemes

It is possible to relate either of the above functions to a fault current iF(t).

Consider Figure 9.1 (b), where an internal fault is assumed to occur at a point F.
If r, is the surge travelling time from R to F and r 2 is the surge travelling
timefrom F to S, then Bergeron's equation for the sections of the line S-F and
F—R can be obtained using eqns. 9.1 and 9.2, respectively, such that
1 1
iFs{t ~ r 2 ) + — yF(/ - r 2 ) = - is{t) + — vs(t)

1 1
"-iFR(t-T2)+— vF(t-T2) = iR(t-Tl-T2)+ — vR{t-rx-r2)
z0 z, 0
Combining these equations and noting that r, + r2 = r, results in
1

Now it can be seen from Figure 9.1 and eqn. 9.3 that
et(t) = iF(t-r2) (9.4)
Similarly it can be shown that
e2(t) = ip(t-rl) (9.5)
It will be evident that the functions £l(t) and £2(t) effectively describe the
current in the fault path in terms of the voltages and currents at the line ends. A
nonzero value for £{(t) and £2(t) effectively signifies that a fault exists between
the measuring points, a fact that can be used to generate a trip signal. The
formation of the signals £j(/) and £2(t) require the transmission of the voltage
and current waveforms between the line ends to form, in effect, a unit protection
scheme. This technique has recently been implemented for plain feeders [3, 4]
and has been extended to Teed power circuits [5]. Further details of the
implementation are given in Chapter 11.

Three-phase lines
The concept of detecting internal faults using Bergeron's equation can easily be
extended to three-phase lines. Consider the three-phase line shown in Figure
9.2. Assume the line is transposed, lossless and has a series per unit length
inductance matrix L and a shunt capacitance matrix C. The line can be
decomposed into three equivalent single phase lines using eqns. 8.23, such that

VM S V l)
' " '' ,9.6,
i.»=<r'v>
As previously explained, the voltage vector vm(t) defines three separate modal or
component voltages, which can be derived from the actual phase voltages
described by the vector vp{t). The component currents im(t) can be similarly
described in terms of the phase currents. The single-phase theory developed
above can then be applied to each component or mode voltage and current
 Travelling-wave protective schemes 151

thereby derived, and the previously derived discontinuity functions of eqn. 9.3
can be written for each mode component. For the kth mode eqn. 9.3 is thus
written as
1

(9.7)
{ k)
2 (t)
(t) == i f ( / -

where Z{ok) is the kth mode surge impedance and x{k) is the kth mode surge
travelling wave time, where A; = 0, 1, 2 for a three-phase line.
Thus the fault detection criteria in a three-phase line can be written as
follows:
• If £{0)(t) = 0 and £{l)(t) = 0 and e{2)(t) = 0, then the line is healthy
• If £{0)(t) ¥=0 or £{])(t) # 0 or £{2)(t) =£0, then an internal fault has occurred.
The physical meaning of £{k) and e^ is that they describe the modal component
values of the current in the fault path, i.e.
£f{t) = if{t-xf), * = 0, 1 and 2
(9.8)
£i*)(/) = ij*>(/-ri* ) ),* = O, 1 and 2

R ^
iR(a) i s (a]
^^ ^

i R (b) is(b)

IR(C) 'S(C) vs(a)

v R (a)

v R (b) v s (b)

v R (c) vs(c)

Figure 92 Distributed three-phase line model

152 Travelling-wave protective schemes

where if\t) is the £th modal component of fault path current, t\k) is the A:th
modal surge travelling wave time from the point F to the terminal S and r ^ is
the kth modal surge travelling wave time from the fault point F to the terminal
R.

9.3 Ultra-high-speed polarity comparison scheme [6, 7]

An ultra-high-speed relay suitable for EHV/UHV applications was developed
by Asea, as a part of a joint American-Swedish research programme. The first
relay developed was implemented using analogue circuits and was designated
'RALDA'; it formed the basis of a number of digital protection equipments
which have emerged since the original research was performed in the late 1970s.

9.3.1 Basic operating principle

The operating principle of the scheme is based on the relative polarities of the
superimposed quantities. As explained in Section 8.3.1, for internal faults the
superimposed current and voltage have different polarities at both ends of the
protected line. Conversely, for external faults the polarities are different only at
one end of the line.

9.3.2 Description of typical implementation [6]

Figure 9.3 illustrates schematically the basic arrangement. Basically, this figure
shows a transmission line which is protected by two RALDA type relays, each
installed at one end of the line. The relays communicate with each other by
means of a communication channel linking the two relays. This enables them to
exchange information about the relative polarities of the superimposed quanti-
ties at each end, which in turn is used to decide whether the fault is internal or
external to the line. The overall arrangement forms a directional comparison
scheme, in which, if both relays detect a fault in a direction looking into the line,
a trip signal is generated. Any one relay identifying a fault looking into either
source thus indicates that the fault is external to the protected line, and in this
case breaker tripping is inhibited. The actual measurements are based on
information derived at single measuring points, i.e. at each end separately, and
the scheme is thus a non-unit measuring technique which is effectively unitised
by employing a directional signalling (binary) channel.
Each relay consists of an analogue interface, steady-state frequency suppres-
sors, amplifiers, directional detectors, logic and drivers and an output interface.
The analogue interface accepts input signals from the current and voltage
transformers and provides the required galvanic isolation and surge immunity.
The steady-state frequency suppressors are designed to extract the superim-
posed current and voltage components caused by the fault. This is achieved by
suppressing prefault components contained in the waveforms at the output of
the analogue interface. The signals are then connected to amplifiers, which in
effect control the basic sensitivity and are designed to achieve a minimum or
 Travelling-wave protective schemes 153

terminal A transmission line terminal B

TC

ryyys nryy\

analogue
interface
analogue ~3
interface _ J E
sss
steady-state
sss
steady-state
frequency suppressors frequency suppressors

i>
amplifiers amplifiers amplifiers amplifiers

directional directional directional! directional

detector detector detector detector

local local
trip block block trip
trip trip
information link

logic and drivers logic and drivers

Tr Tr

output output
interface interface

tripping and R = receiver tripping and

signalling Tr = transmitter signalling

Figure 9.3 Block schematic diagram for RALDA system [6]

154 Travelling-wave protective schemes

maximum reach along the line, depending on whether an over-reach or under-

reach operation is desired. The signals are then fed to directional detectors
where the direction of the fault is determined by comparing the relative
polarities, i.e. the signs of the superimposed voltages and currents. The
detectors then initiate a tripping direction signal T if the signs of i and v are
opposite to each other. The other set of detectors seen on the block diagram are
used in a parallel mode to enhance the dependability and speed of operation.
Each of these units has an under-reach setting, which enables direct tripping of
the local circuit breaker for close-up faults. This mode of operation is
intrinsically selective and does not require time co-ordination. This makes it
particularly valuable for ultra-high-speed clearance of close-in faults. Typical
operating times in this mode of operation can be as low as 2-3 ms.
The output signal from the directional element of each phase is processed in a
logic circuit, which has been arranged to perform several functions. First, it
determines the sequence in which trip and block signals are generated and,
directly on fault inception, it establishes whether the trip signal (T) was sensed
before the block signal (B) or vice versa. This avoids the confusion that could be
caused by multiple reflections of the incident waves. It also determines whether
the fault is a single- or multi-phase fault and performs the appropriate phase
selection as required in single-phase tripping schemes. In addition, it co-
ordinates the local relay decision with the remote end relay via the communica-
tion link. Because each relay terminal has a local phase selection capability, the
signal transmitted over the communication link is either a trip or block signal.
The signals from each logic circuit are connected to the output interface unit,
which includes solid state tripping relays and electromechanical signal relays
with corresponding flag indicators. It also has the facility to display signal
tripping, blocking, phase selection and mode of operation, as well as alarm
functions.
To provide full galvanic isolation between the relay circuitry and the
auxiliary power source and to suppress any surges imposed on the DC input,
the auxiliary power is supplied through a DC/DC converter.
As far as the design of the communication link is concerned, the most
important factor to be taken into account is the speed. This is determined by the
clearing time requirements for remote end faults, and an all-inclusive relaying
time of approximately 8 ms for remote-end faults has been specified for this
scheme.

9.4 Ultra-high-speed wave differential scheme [8, 9]

9.4.1 Operating principles
The basic principle of the wave differential scheme is that the direction to the
fault can be determined by the sequence in which Sfb S^R, k = Q, 1, 2 exceed
given setting levels (see Section 8.3.2), where S\% S{$ are defined by eqns. 8.48.
 Travelling-wave protective schemes 155

For convenience they are rewritten below for each mode:

5.4.2 itarzV description of the scheme

In common with the RALDA system, the earliest work was done using a largely
analogue implementation, and Figure 9.4 shows a block schematic diagram of
an analogue prototype scheme. The currents and voltages derived from primary
transformers are taken to their respective mixing circuits. In these circuits the
currents and voltages are decomposed into their respective modal time
variations according to eqns. 9.6. The modal currents are passed through surge
replica circuits, where each modal component is multiplied by its corresponding
replica surge impedance, which in turn results in a signal waveform R<pi$(t)9
k = 0, 1, 2. The modal voltage variations v$ and those formed from the modal
current components (R}pi]S>\t)) are then mixed in the signal-mixing circuit/
analogue interface block to form the composite signals vff — RtfHffit), k =
0, 1,2. It is important to note that at this point in the process the composite
signals contain both steady-state and superimposed quantities.
Each composite signal is passed through a superimposed component extrac-
tion circuit, where the superimposed components are extracted from the total
time variation of the composite signals; Figure 9.5 (a) shows an analogue
technique that was used to achieve this. Fig 9.5(b) illustrates the steady-state
and the superimposed waveforms associated with an arbitrary chosen total
waveform, from which it is evident that the superimposed component can be
obtained over a limited period of time by delaying the total waveform by an
integer multiple of the period T= 1 /f0. The use of a time delay of one cycle of
nominal system is generally satisfactory, and Figure 9.5(c) shows the circuit
waveforms that are often obtained. In particular it shows how the circuit of
Figure 9.5(a) produces the superimposed component for one power frequency
period (7) after the latter becomes finite.
The composite signals due to the superimposed quantities are then taken to
the sequence detector block, which examines the direction of the fault in
accordance with the sequence in which the discriminant signals of eqn. 9.9
exceed a threshold ( ± Vs), as explained in Chapter 8. Figure 9.6 illustrates in
more detail the basis of the operation of the sequence-detector circuits for the
mode 1 (k = 1) channel following forward and reverse faults.
Each level detector in the circuit of Figure 9.6(a) is arranged to give an output
when the magnitude of the input signal exceeds a preset pickup or setting
156 Travelling-wave protective schemes

current- voltage- similar arrangement

mixing mixing to end R
circuits circuits

surge
replica
circuits

signal-mixing-circuits/analogue interface

to similar arrangement to similar arrangement

for mode -0 signals £ for mode -2 signals

superimposed
trip outputs from component block outputs from
0 and 2 channels circuits modes 0 and 2 channels

sequence
detector
(mode-1)

similar
)• arrangement
to end R

blocking relay

Figure 9.4 Block schematic diagram for wave-differential scheme

 Travelling-wave protective schemes 15 7

R«(t-TVR0«viR«(t = T)
(k) (k) (k)
v R (t)-Ro iR (t) time delay T
(BBD)

differential
amplifier

steady-state
component

true superposition
component

time

Figure 9.5 Basis of arrangement for extracting superimposed components

(a) circuit arrangement
(b) relationship of steady and superimposed components to total signal
variation
(c) waveform illustrating circuit action
158 Travelling-wave protective schemes

7/777

Figure 9.6 Sequence detector arrangement and function

(a) schematic arrangement of mode 1 channel
(b) circuit function for forward faults
 Travelling-wave protective schemes 159

rui OIL
—LTU—inr 77777 (ii)

_HJi n_n_ T L = 60ms

7777 (vii)

_iui n_n_
(x)
77/77
rL = 60ms (v)
77777

Figure 9.6 Sequence detector arrangement and function (continued)

(c) circuit function for reverse faults

voltage Vs. On the other hand, an input to the pulse-generating circuits PG1 or
PG2 causes their output to fall to zero for a preset time TL. The latter time
determines how long the detector is latched following a forward or reverse fault,
thus ensuring that any subsequent change in the input signals is ignored.
Similar considerations apply to PG3, but this circuit is arranged so that its
output is normally zero and rises for a time of 60 ms following an input. The
tripping signal has to be delayed to enable the local blocking relay to operate on
receipt of a carrier blocking signal generated by the relay at the other end of the
line following an out-of-zone fault. This is achieved by the inverting delay
circuit which delays the trip signals by the required time.
160 Travelling-wave protective schemes

9.4.3 Digital implementation of wave differential scheme [9]

The simple sequential threshold decision technique adopted in the analogue
UHS wave differential scheme provides only a limited degree of security and
dependability. Particular problems emerged largely from difficulties in guaran-
teeing satisfactory operation when the relaying signals were corrupted by noise
or affected by primary devices in some very-high-noise substation environ-
ments. To improve performance and take advantage of developments in digital
technology, the UHS wave differential scheme and commercial equipments
derived from them have been engineered using digital signal processing
techniques.

Operating principle
The operating principles adopted for the digitised UHS scheme are basically
the same as those used in the analogue UHS scheme. For ease of explanation,
consider the superimposed single-phase equivalent circuit shown in Figure 9.7.
The composite superimposed signals used are those defined in eqns. 8.44,
rewritten below. As also explained previously, the actual signals used employ
modal components but for ease of explanation only the equivalent single phase
form of signals repeated in eqn. 9.10 will be considered.
SlR(t) = AvR(t)-RQMR(t)
(9.10)
S2R(t) = AvR(t)+R<£iR(t)
Signal magnitude comparison criteria were used to enable the directional
decision to be based on several consecutive samples of the composite signals of
eqn. 9.10, thereby improving relay security. This is desirable because, for a
reverse fault behind the relay location in Figure 9.7, the simple signal criterion
of eqn. 9.11 only holds for twice the wave transit time (l/c) between the line
ends. In the case of a forward fault at a distance xF from the relay location
considered in Figure 9.7, the signal criterion of eqn. 9.12 holds only for so long

source source
R relay location fault S
AiR(t) F
l
Av R (t) v F (t)

XF

Figure 9.7 Simple superimposed component faulted circuit model

 Travelling-wave protective schemes 161

as reflections from within the source network R take to return to the relay
location.

\Sw{t)\>\SXR{t)\ (9.11)
(9-12)

It will thus be apparent that in some circumstances the discriminative criteria

eqns. 9.11 and 9.12 are valid only for relatively short periods after fault
inception. However, they can be made to hold for extended periods of time by
decreasing the influence of travelling wave components. This is illustrated in
Figure 9.8, which shows the superimposed voltage and current due to a forward
solid fault on the system shown in Figure 9.7 but with the line being represented
by lumped elements, i.e. travelling waves are neglected. If the distributed
nature of the line is ignored, as is often the case in practice, then the resulting
travelling-wave components would simply be added to the superimposed power
frequency and exponential transient components. Figure 9.8 shows that, when
the travelling waves are absent, the forward directional criterion of eqn. 9.12
would hold for approximately one half cycle and one quarter cycle of the power
frequency for faults at zero and maximum points on the wave, respectively.
Specially designed digital filters are therefore used to decrease the influence of
travelling-wave components after the initial post-fault period, thereby extend-
ing the time for which the discriminative criteria hold.
Variable thresholds, against which the filtered directional wave signals are
compared, are used. These are adjusted in line with the prevailing noise in
substation environments. This approach allows a sample-by-sample counting
regime to be implemented, which also takes account of any momentary high-
level noise bursts above the basic threshold, and therefore improves relay
integrity. It also improves the response where significant signal components
introduced by capacitor voltage transducer transients are generated and
ensures that fast relay recovery is achieved.

9.4.4 General description of the digital relay

Figure 9.9 shows the block diagram of the digital form of the wave-differential
relay. It can be seen that the design of the relay is based on using detectors for
two modes of propagation only, thereby simplifying the hardware requirement.
The primary transducer signals are converted, via the relay interface circuits,
into analogue form for electronic pre-filtering and modal mixing before
conversion to digital form for subsequent digital filtering and decision process-
ing. After the modal mixing analogue stage, the two mode voltage and current
signal pairs are processed as far as the directional decision stage in independent
channels, and the resulting forward and reverse directional decisions are
combined as shown in a common scheme logic processor. Adjustable gains
within the analogue voltage and current channel facilitates the sensitivity
 a3

fault fault fault s

Av R (t)

Figure 9.8 Typical superimposed voltage and current for lumped parameter system model
(a) fault at zero voltage
(b) fault at voltage maximum
(c) fault at 145° after voltage zero
 Travelling-wave protective schemes 163

-m line

v a ,v b .v c

i a > >b> >c

w. V
forward/reverse
detection signals

scheme logic

breaker
trip TT remote relay
block signals

-O
-0-0 receive send

Figure 9.9 Digital relay block diagram

164 Travelling-wave protective schemes

selection for particular applications. The arrangement of Figure 9.9 forms the
basis of the digital algorithmic relay developed by GEC Alsthom
Measurements, type designated LFDC.

9.5 Discriminant function based scheme

9.5.1 Operating principles [10, 11]
The operating principle of this scheme is based on using discriminant functions
(see Section 8.5). This can best be explained with reference to Figure 9.10 where
the table shown is constructed by calculating the modal discriminant functions
using eqns. 8.71 and 8.72 for different types of fault. The values calculated are
normalised with respect to F| MS , i.e. the square of the operating voltage, to
produce 0s and Is. This table is based on the Karrenbauer transformation
described by eqn. 8.28. It will be noted that this table consists of three main
blocks marked phase 'a', 'b' and V . The values tabulated under these blocks
are obtained by assuming that the basis of the transformations is phase 'a', 'b'
and V , respectively. The same procedure could be applied using the Wedepohl
and Clark transformations described in eqns. 8.27 and 8.29. Figure 9A0(b)
shows a flowchart that explains how the phase selection and fault classification
can be achieved using modal discriminant functions. This can be achieved
using the components D%, D%, Z)f, D%, D%, D%, where D{k) is the * modal
discriminant function, and the first subscript is used to indicate whether it is a
forward (F) or backward (B) function. The second subscript is used to refer to
the phase on which the basis of the transformation is assumed. If the second
subscript is omitted, the discriminant function is due to a three-phase fault
condition.
As shown in the flowchart of Figure 9.10(£), the forward discriminant
components D^a and D§\ are first compared with a threshold value. A forward
fault is indicated if either of the two components exceeds the threshold. Under
this condition, the phase selection and fault classification are determined from
the four forward components listed above and in accordance with the logic
shown on the right-hand side of the flowchart. However, if none of the forward
discriminant components D^a or D$b exceeds the specified threshold, the
backward components D$a and D$a are checked. If either of them exceeds the
threshold a backward fault is indicated, otherwise the system is considered
sound. In the latter case the above procedure is repeated for the next set of
samples. The measurements must be prefiltered, to avoid undue sensitivity to
low levels of noise.

9.6 Superimposed component trajectory based scheme [ 12]

9.6.1 Basic principles
The direction of fault can be determined from the sense of rotation of the
trajectories resulting from the superimposed quantities, in such a way that
 Travelling-wave protective schemes 165

fault _ ^ L-G L-L L-L-G

3LS
basis D's a b c a-b b-c c-a a-b b-c c-a
D (0) 1 1 1 0 0 0 1 1 1 0
Ph
D (D 1 1 0 1 1 1 1 1 1 1
"a"

D (2)
1 0 1 1 1 1 1 1 1 1

D(0) 1 1 1 0 0 0 1 1 1 0

Ph 0 1 1 1 1 1 1 1 1 1
"b"
D(2) 1 1 0 1 1 1 1 1 1 1

D(0) 1 1 1 0 0 0 1 1 1 0

Ph D d) 1 0 1 1 1 1 1 1 1 1
"c"
D (2)
0 1 1 1 1 1 1 1 1 1

no
D(i) F l b orD(i) F l a > £ \

no^
V L-L or
3L's fault K D(0)Fla<£

no disturbance backward fault

trip "a"

trip "b"

trip "c"

L-L-G
fault

Figure 9.10 Fault detection, directional discrimination, phase selection, andfault classifi-
cation based on the Karrenbauer transform [11]
(a) truth table
(b) process flow chart
166 Travelling-wave protective schemes

forward faults give rise to a clockwise direction of rotation, while reverse faults
give rise to trajectories with anticlockwise rotation.
9.6.2 Sense of trajectories versus fault direction
It has been shown in Section 8.3.4 that the behaviour of the superimposed
voltage and current at the relay point can be described by an elliptical
trajectory in the AvR versus RAiR plane, where R is an arbitrary scale factor.
The trajectory itself is mathematically represented by eqn. 8.55, rewritten
below. (Note that subscript R is omitted in the following analysis.)
Av2(t) [RAi(t) - RAI cos Of
l (9J3)
[RAIf
Figure 9.11 shows the elliptical trajectories due to forward and reverse faults
with respect to a relay located at the R end of the line R-S. Figure 9.11 (a)
illustrates the superimposed component trajectories due to forward faults.
These trajectories possess a clockwise sense of rotation and they depart from the
Av axis, either into quadrant II or IV, depending on whether the fault occurs on
the positive or negative half cycle of the prefault voltage waveform.
On the other hand, Figure 9.11 (b) illustrates the trajectories due to a reverse-
fault condition. Here it can be seen that the trajectories rotate in an
anticlockwise direction and initially enter quadrant I or III, again depending
on which half cycle of the voltage waveform the fault occurs in. Therefore, a
basic technique to determine the direction of the fault using the above
properties of the superimposed component trajectories is to introduce threshold
boundaries in the Av—RAi plane. Once the trajectory has crossed the boundary,
the relay may reset or adjust itself to the postfault sinusoidal voltages and
currents; a typical boundary is shown in Figure 9.12. A simplified method of
boundary checking involves taking linear combinations of Av(t) and RAi(t) and
checking against fixed values. For example, if the linear combination of
RAi(t)-Av(t) first reaches some threshold value, say +£"0 or — Eo, a forward
fault will be indicated. Mathematically this is expressed as
RAi(t)-Av(t) = ±E0 (9.14)
However, if the linear combination of RAi(t) 4- Av(t) first reaches the threshold
values, then a reverse fault will be indicated. More explicitly this can be
expressed as
RAi(t) + Av(t) = ±E0 (9.15)
Boundaries such as hyperbolas, circles, ellipses or piecewise linear curves can
also be used.

9.6.3 Extension of trajectories approach to signals including travelling-wave

components
The superimposed component trajectories which have been discussed so far are
confined to quasi-steady-state conditions. The same principles can, however, be
 Travelling-wave protective schemes 16 7

RoAi

Figure 9.11 Elliptical trajectories due to

(a) forward faults
\b) reverse faults
168 Travelling-wave protective schemes

forward Mult

A\ c^s (9)

Figure 9.12 Typical boundary used to determine fault direction in trajectory-based relay
[12]

extended to transient conditions and particularly to the case when travelling-

wave components are present in the signals.
It will be recalled from the analysis of a single-phase transmission line having
constant distributed inductances and capacitances (see Section 8.2.1) that the
voltage and current is made up of forward and backward travelling waves. Now
let us consider the superimposed signals resulting from a change in voltage vF
due to a fault at point F on a distributed single-phase line such as that shown in
Figure 9.13. If we denote the instantaneous values of the forward and backward
travelling waves (as seen by the relay) by f(t) and b(t)9 respectively, the
superimposed voltage and current Av(t) and &i(t) at the relay point can be
expressed by solving the resultant wave equation using the D'Alembert
approach:

(9.16)
1

where ZQ is the surge impedance of the line.

By solving eqns. 9.16 for/(/) and b(t), we obtain
f(t)=Z0Ai(t) + Av{t)
(9.17)
b(t) = ZoAi(l)-Av(t)
A logical approach to assess the location of the fault with respect to a relay
location is to detect which of the two waves first reaches a given threshold
constant Eo. Therefore if the wave signal b(t) associated with the backward
travelling wave reaches the threshold value ±J£o> t n e n a fault in the forward
 Travelling-wave protective schemes 169

point of t=0
local reflection remote
source^ R 3 source

Figure 9.13 Travelling waves generated by a fault

internal fault
external fault
Zero fault resistance assumed [12]

direction is assumed. This condition can be expressed mathematically as

Z0M(t)-Av(t) = ±E0 (9.18)
On the other hand, if the wave signal/(/) associated with the forward travelling
wave increases beyond the absolute value of EQ> a reverse fault will be indicated.
This condition is given by the following equation:
±E0 (9.19)
The operating conditions described by eqns. 9.18 and 9.19 are identical to eqns.
9.14 and 9.15, which describe the slope boundaries previously discussed for the
assumed quasi-steady-state condition provided that the parameter Zo is chosen
as Z0 = R. This is achieved in a practical implementation by using a current
mimic impedance RQ in deriving the signal component due to the superimposed
current At(/).
170 Travelling-wave protective schemes

9.7 References
1 TAKAGI, T. et al.: 'Fault protection based on travelling wave theory, Part I: theory',
PES Summer Meeting, Mexico, 1977, Paper A77, 750-753
2 TAKAGI, T. et al.: 'Fault protection based on travelling wave theory, Part II:
sensitivity analysis and laboratory test', PES Winter Meeting, 1978, Paper A78, 220-
226
3 AGGARWAL, R.K., and JOHNS, A.T.: 'A differential line protection scheme for
power systems using composite voltage and current measurements', IEEE Trans.,
1989, PWRD-4, pp. 1595-1601
4 AGGARWAL, R.K., and JOHNS, A.T.: CA new differential protection scheme for
power systems using composite voltage and current measurements', 23rd UPEC,
1988, Paper A3.4
5 AGGARWAL, R.K., and JOHNS, A.T.: 'New approach to teed feeder protection
using composite current and voltage signal comparison', Developments in power
system protection (IEE Conf. Publ. 302, 1989), pp. 125-129
6 CHAMIA, M., and LIBERMAN, S.: 'Ultra-high speed relay for EHV/UHV trans-
mission lines—development, design and application', IEEE Trans. 1978, PAS-97,
pp. 2104-2116
7 YEE, M.T., and ESZTERGALYOS, J.: 'Ultra high speed relay for EHV/UHV
transmission lines—installation-staged fault tests and operational experience', ibid,
pp. 1814-1825
8 JOHNS, A.T.: 'New ultra-high-speed directional comparison technique for the
protection of EHV transmission lines', IEE Proc. C 1980, 127, pp. 228-239
9 JOHNS, A.T., MARTIN, M.A., BARKER, A., WALKER, E.P., and CROSSLEY,
P.A.: 'A new approach to EHV directional comparison protection using digital signal
processing techniques', IEEE Trans. 1986, PWRD-1, pp. 24-34
10 MANSOUR, M.M., and SWIFT, G.W.: 'A multi-microprocessor based travelling
wave relay—theory and realization', ibid, pp. 273—279
11 MANSOUR, M.M., and SWIFT, G.W.: 'Design and testing of a multi-
microprocessor travelling-wave relay', ibid, pp. 74—82
12 VITINS, M.: 'A fundamental concept for high speed relaying', IEEE Trans. 1981,
PAS-100, pp. 163-173
 Chapter 10
Digital differential protection of
transformers

10.1 Introduction
This Chapter gives a brief general review of the principles of transformer
differential protection. This is followed by an explanation of the application of
digital techniques and the algorithms that have been developed specifically for
application to transformer protection. The algorithms covered include finite-
duration impulse (FIR) filters, least-squares curve fitting, the digital Fourier
algorithm and the flux-restrained current differential algorithm.
Finally, the basic hardware arrangement for implementing digital techniques
to the protection of transformers is described. It is, however, important to note
that closely similar techniques can be applied to the protection of generators,
although, in this case, the transformation ratio of currents is the same on each
side of the protected zone.

10.2 Principles of transformer protection [1-4]

10.2 J Basic principles
The most commonly encountered transformer protection arrangement is based
on the differential current principle. This can be illustrated by reference to
Figure 10.1, in which the primary and secondary currents (lp and Is) are
compared after being reduced by their corresponding current transformers. The

transformer

Figure 10.1 Basic principle of transformer differential protection

172 Digital differential protection of transformers

primary and secondary current transformers are connected such that ideally,
under healthy conditions, only circulating currents flow and the differential
signal ID is zero. If an internal fault occurs within the transformer, the balance
between the primary and secondary currents is disturbed and a differential
current signal (//> = Ip — Is) then causes the relay to operate.
10.2.2 Biased differential relaying
If the transformer is equipped with a tap changer, the imbalance between the
primary and secondary currents introduced by the variation of the turns ratio
can be great enough to cause malfunction during external fault conditions.
Imbalance under healthy conditions can also be caused by mismatch between
the current transformers or by saturation. It is therefore important that this
situation should be taken into account in the engineering of transformer
differential protection systems.
A common way of overcoming such problems is to bias the operation of the
relay by deriving a biasing signal equal to the sum of the currents measured on
each side of the transformer. This is illustrated in Figure \0.2(a) together with a
typical operating characteristic (Figure \0.2(b)). It is evident from Figure
\0.2(b) that the differential current required to operate the relay increases with
the bias current. Thus the stability of the relay under external fault conditions is
ensured. For example, if a heavy through fault (healthy) condition exists, the
differential current may take a finite value that is nevertheless much smaller
than the associated bias current. The restraining characteristic of Figure \0.2(b)
is arranged to allow a significant level of differential current to occur without
causing tripping under healthy conditions. Conversely, a faulty condition
causes the bias current to be relatively small, on account of the reversal of the
current measurand Is, thus ensuring that tripping occurs.
In some cases the operating current is expressed as a percentage of the
restraining current, and such relays are commonly known as percentage-biased
differential relays. Many practical designs try to derive differential and bias
signals proportional to the fundamental or power frequency components of
measured current, which is why in Figure \0.2(b) the measurands are
subscripted.
10.2.3 Harmonic-restrained differential relay [5, 6]
The differential protective scheme described above is generally very satisfactory
under normal operating conditions, but is prone to false operation during
energisation of the transformer. Under this condition, an inrush current flows
only through one winding (the winding being energised), which consequently
appears to the relay as an internal fault.
This problem can be overcome by using the fact that inrush currents usually
contain harmonics. Of these, the second harmonic is usually predominant
under all energisation conditions. The protection can therefore be designed so
that it is prevented (restrained) from tripping if the magnitude of the second-
harmonic component of operating current is greater than a certain prescribed
 Digital differential protection of transformers 173

percentage of the fundamental component. Fifth harmonics, which are gener-

ated under over-excitation conditions (caused, for example, by temporary
overvoltages) are also used in some equipment to prevent the relay from
tripping under energisation conditions.
It will be apparent that harmonic-restrained current differential transformer
protection basically uses signals formed from fundamental (or power frequency)
and harmonic components. The power-frequency components predominate
when an internal fault occurs, and the harmonic and bias signals are used to
restrain the protection during through faults, switching operations and during
over-excitation conditions.
With reference to Figure 10.2(a), the differential current ID is commonly
formed from the fundamental frequency components of the primary and
secondary currents lXp and Ils, respectively, such that

A A A A
V V
bias signal
bias signal component (ls)
component (IP)
differential signal

relay operates

relay restrains

restraint/bias current,
=
lm Mp"~Ms

Figure 10.2 Biased differential relay

(a) basic arrangement
(b) typical operating characteristic
174 Digital differential protection of transformers

The restraining signal (Im), on the other hand, is commonly formed from the
fundamental components of the primary and secondary components such that
4i=/M+4 (10.2)
The other restraint signals are the second-harmonic restraint Im and fifth-
harmonic restraint current IR5, which are commonly formed from the magni-
tudes of differential currents of any second and fifth harmonics detected. A
commonly used technique involves summing any such harmonic components in
each phase:
/«=|/i».l + |/i»l + |/OTe| (10.3)
and
/*5=|/zJ + | / / J + |/z>5,l (10.4)
where ID2af Imif and ID2c are the differential currents due to second-harmonic
components in phases 'a', 'b' and V, respectively, and ID5ay Imb and ID5c are the
corresponding differential currents due to fifth-harmonic components.
Tripping and restraining signals can be derived using analogue techniques
(as in the case of electromagnetic relays or solid-state relays) but this Chapter
deals with the derivation and use of such signals using digital techniques.

10.3 Finite-duration impulse response filter based

algorithms [7, 8]

These algorithms are based on using finite-duration impulse response (FIR)

filters to estimate the magnitude of the fundamental and second-harmonic
components. The estimates are then used to calculate the ratio of the magnitude
of second harmonic to fundamental. If this ratio is found to be greater than a
certain value, an inrush current condition is assumed. This algorithm is often
designed to avoid multiplication and division as these operations are rather
unwieldy to perform at both software and hardware levels.

103A FIRfiltercharacteristics
The filters used are assumed to be characterised by means of finite-duration
impulse responses of duration equal to one power-frequency cycle of period T
and having values of +1 or — 1 at any instant during that period. The choice of
± 1 for the impulse response avoids the need tofindthe responses of thefiltersto
arbitrary inputs and eliminates the time-consuming multiplications that would
otherwise be required, although in many modern DSP processors this is not
necessary.
To estimate the magnitude of fundamental and second-harmonic compo-
nents of the input, four filters are required: two for the fundamental and two for
 Digital differential protection of transformers 175

the second-harmonic component. Their impulse responses are S{(t), C{(t), S2(t)
and C2(t) (Figure 10.3), which in turn are defined by the following equations:

(ia6)

, 7/2^/^37/4
-1 774^/^772, 3 7 / 4 ^ / ^ 7
+ 1 0^/^7/8, 37/8^/^57/8,
, 57/8=^/^

It will be apparent that £,(/), C,(/) are impulse responses for the sine and cosine
parts of the fundamental component, and S2(t), C2(t) are impulse responses for
the sine and cosine parts of second-harmonic components. Eqn. 2.73 can be
applied to eqns. 10.5-10.8 to find the system frequency responses of the four
filters, which are consequently given by

Fsl(o>) =£• c-^coS(^Pj - 1^ (10.9)

^ - 2 ™(^

=1 e-J">™(coS(^Pj - 2 c o s ^ + 1^ (10.11)

(^)) (10.12)

where T=2tt/(DQ is one period of the system frequency^.

The amplitudes of the above frequency responses plotted against the ratio
O)/O)Q are shown in Figure 10.4.
10.3.2 Extraction of fundamental and second-harmonic components
To extract the fundamental component of a current, it is necessary to determine
the responses of the filters to the transformer current. The responses of the
filters 1,2,3 and 4 to transformer current after one cycle of power frequency
(t = 7) can be found by convolving the impulse responses of thesefilterswith the
transformer current (as an input to the filters) and evaluating at time t=T.
As always in digital protection, the current inputs to thefiltersare in the form
of samples, and since the impulse responses of the FIR filters have the values of
either +1 or —1 at any instant of time (as defined by eqns. 10.5-10.8), the
multiplication operation required by the convolution process is simply a sign
176 Digital differential protection of transformers

T/2 3T/4

-1

-•t
T/4 T/2 3T/4 T

-1

S2(t)

T/2

-1

C2(t)
i
1

>
T/4 T/2 3T/4 1r

-1

Figure 10.3 Impulse response of fundamental and second harmonic FIR filters
 Digital differential protection of transformers 17 7

1 2 3 4 5 6
a

0.5

1 2 3 4 5 6

0.5

IF S2 ((o)l

2 3 4 5
c

0.5
IFC2(©)I

1 2 3 4 5 6
d

10.4 Magnitudes of the frequency responses of fundamental and second-harmonic

filters
(a) filter 1
(b) filter 2
(c) filter 3
{d) filter 4
178 Digital differential protection of transformers

change. Thus the outputs of filters 1, 2, 3 and 4 are obtained by summing the
samples of their inputs over one cycle. However, the signs of these samples must
be modified in accordance with the impulse responses of the filter under
consideration.
Assume N is the number of samples per cycle of the current i(t) and is chosen
as a multiple of eight. In this case, the time between successive samples is
At = 2Jt/(N(o0) and ik—i{tk) is the Mi sample at any time t = kAt. The result of
time-discrete convolution of the samples ik with the impulse responses defined
by the previous equations will then be given as

(10.13)

(10-14)
k=l

JV/4

(10.15)
k=\

A78

(10.16)

103.3 Discrimination between inrush and internal fault currents

The criteria used to distinguish between inrush currents and currents due to
internal faults is based on evaluating the ratio £, which is calculated from the
ratio of the larger of the two components of each pair of filter outputs as
described by eqns. 10.13-10.16 such that
max(|.S2l, |C 2 |)
(UU7)

An internal fault, as distinct from a current inrush condition, is characterised by

a high value of fundamental component of differential current and a relatively
low value of second harmonic. Conversely, inrush conditions are associated
with relatively high second harmonic current components. It has been sug-
gested [7] that the values of e for the detection of internal faults are

e ^0.093 X/R=\0
0.054 X/R = 20
where XIR is the system reactance-resistance ratio.
 Digital differential protection of transformers 179

A typical threshold value € 0, which is convenient from a computational point

of view, is £o = O.125. In such a case, any value e which is greater than 0.125
would be taken to indicate an inrush condition and would inhibit operation of
the protection.

10.4 Least-squares curve fitting based algorithms [9, 10]

Least-squares (LSQ) curve fitting methods have been discussed in detail in
Chapter 6. These methods are basically used to extract the fundamental and/or
any other harmonic components. Since the relative values of second-harmonic
and fundamental components of the differential current play an important role
in the transformer differential protection, LSQ techniques can be used either to
detect the second harmonic in an inrush current (which is later used to block
the differential protection) or to find the ratio between the fundamental and
second-harmonic components of the differential current, which in turn is used
to differentiate between inrush current and internal faults.

10.4.1 Basic assumptions and algorithm derivation

Since inrush currents can in fact contain significant harmonics up to the fifth,
the differential current can be described by a waveform containing a decaying
DC component, fundamental and harmonic components up to the fifth order:

Imcos(mco0t+dm) (10.18)

where r is the time constant of any decaying DC component.

If the decaying DC component is approximated by the first two terms of its
Taylor expansion, then the current sample i(tk) measured at tk can be expressed
as
5 5

i(^) = /o-(/o/T)/* + ^ ( 4 cos 0m) cos moy* + X (7«« s i n 0 «) s i n m(

°oh
m=l m=\

(10.19)
or

with

02 = 4 * 2 =-/ 0 /r
aa+2 = cos na)otk) xn+2 = /„ cos 0n
an+7 = sin n(O0tk) xn+7 = In sin 6n
180 Digital differential protection of transformers

To solve for the unknowns xn(n= 1, . . . , 12), m equations can be constructed

from N current samples. The resulting equations can be written in matrix form:

A J£_,= « (l0-20)
JVX12 12X1 JVxi

or

X=Bi (10.21)

where B = {AT • A}'1 • AT is the pseudoinverse of A, and AT is the transpose of

matrix A (see Section 2.5.1).
It follows that the real and imaginary parts of the fundamental and second
harmonic can be calculated from eqn. 10.21:
N
Ix cos Qi^Xs^^ b(3, n)in

N
= :=
Ix sin 0 i * 8 ^ . b(8, n)in

(10.22)
I2 cos 82 = x4 = > £(4, n)L

= x9 = ^ b(9, n)in

where xk is the £th element of vector X and b(k, n) is the element of the A:th row
and nth column of matrix B.
Using the foregoing equations, the amplitude of the fundamental and the
second harmonic can be calculated as

In = V(Ia cos dnf+{Itt sin 0n)2 0=1,2 (10.23)

Similar techniques can be applied in extracting, where necessary, the fifth-

harmonic components.

10A.2 Basis of discrimination between inrush and internal fault currents

Discrimination between an inrush and an internal fault condition is often based

on an inrush-detection algorithm. This algorithm is based on comparing the
second-harmonic and fundamental components / 2 and Il9 which are present in
the differential current. The values of these currents, determined according to
 Digital differential protection of transformers 181

x4

Figure 10.5 Schematic representation of the determination of the second harmonic (SH)
ratio
in is the wth sample, b(k, n) is the element of kt\i row and wth column of
matrixB (see eqn. 10.21) and 2 is the sum from n=\ toN, i.e. the number
of samples per fundamental period

eqn. 10.23, are then used to calculate what is called the 'second harmonic ratio',
which is defined by

SHR = - (10.24)

If the SHR is greater than a set value, an inrush current condition is assumed
and tripping is prevented; otherwise an internal fault condition is assumed and
a tripping signal is issued to isolate the transformer.
Figure 10.5 shows a diagrammatic representation of the four digital filters
used to derive the real parts, i.e. x3 and #4, and imaginary parts (x8 and #9) of the
fundamental and second-harmonic components used in calculating the SHR.
The frequency characteristics of the four filters are shown in Figure 10.6 where
it can be seen that the unwanted harmonics are filtered out.

10.5 Fourier-based algorithm [11]

10.5.1 Filtering of harmonics
This algorithm is based on the fact that the fundamental, second- and fifth-
harmonic components contained within an inrush current can be digitally
extracted using the Fourier approach given in detail in Section 5.2.1.
182 Digital differential protection of transformers

Now assume (as previously) that the current waveform is sampled N times
per period of the fundamental, and let the samples be denoted by ik=i(kAt).
The real and imaginary parts of the nth harmonic (an and bn) can be found by

1 -

Ix3(co)l

IX4(G))l

Figure 10.6 The magnitudes of the frequency responses of digital harmonic extraction
filters
(a) xs(a>)
(b) xB(a>)
(c) x<(a>)
(d) x9(a>)
 Digital differential protection of transformers 183

using eqns. 5.5 and 5.7. In terms of current samples starting at the rth sample, an
and bn can then be expressed as

"1 h
2 J*N~l /2jtk\
w= f sin
i v 2 * «hr
(10 26)
-

The magnitude |/£r)| of the wth harmonic and its phase angle 6^ can thus be
calculated as explained previously. Thus

0W = t a i T ' ^ (10.27)

For the transformer protection application, n takes the values of 1, 2 and 5 for
fundamental, second and fifth harmonics, respectively. The result can be
updated iteratively as each new sample becomes available. This is done by
dropping the earliest sample and adding the new sample:

r-ir] cos n(-^pj (10.28)

2nr
A r i ) = *<" + - [iN+r-ir] sin nl — \ (10.29)

where ir and %+r are the oldest and newest samples, respectively.
Having determined the fundamental, second- and fifth-harmonic compo-
nents, the transformer protection is then implemented following the principles
outlined in Section 10.2.3.
Other techniques have been reported in the literature to extract fundamental
and other harmonic components [11]. These techniques give closely similar
performance and include the use of rectangular transforms [12], Walsh
functions [13] and the Haar function [14].

10.6 Flux-restrained current differential relay [15, 16]

This algorithm basically uses the flux-current relation of the transformer to
obtain the restraint function. If the flux could be estimated correctly, it would
provide a sound basis for detecting over excitation as well as magnetising inrush
conditions. An important feature of this technique is that it requires fewer
computations than that based on the Fourier analysis.
184 Digital differential protection of transformers

Figure 10.7 Two-winding single-phase transformer

10.6.1 Development of the algorithm

Consider a simple two-winding tranformer (basically the same analysis can be

applied to a multi-winding transformer) such as that shown in Figure 10.7. It
will be assumed that winding resistance is negligible. The relation between the
primary applied voltage vp, the primary current ip and the mutual flux linkages
ip of the transformer is given by

(10.30)
dt dt

where Lp is the leakage inductance of the primary winding. By rearranging eqn.

10.30 and integrating from tx to t2, we obtain the flux linkages at these times
such that

f'2 (10.31)
J /,
Applying the trapezoidal rule to the integral part of eqn. 10.31 results in
1
(10.32)

If the voltage and current waveforms are sampled such that the sampling
interval is equal to A/, then at the kth sample, eqn. 10.32 can be expressed using
sample notation as

(10.33)

where iPt ky vfiy k are the A:th samples of the primary current and voltage.
Eqn. 10.33 can therefore be used to calculate the mutual linkage flux tp of the
transformer. On the other hand, the differential current can be calculated from
 Digital differential protection of transformers 185

the primary and secondary currents. Thus at time th the differential current idk
is given as:

where iPt k is as defined previously and is k is the kth sample of the secondary
current. However, from transformer theory, the differential current idk is equal
to the magnetising current of the transformer. Therefore, if it is assumed that
the flux linkage computed by eqn. 10.33 is a true representation of the actual
flux in the transformer, the samples of the differential current and flux linkage
(idh \pk) are expected to fall on the open-circuit magnetising curve of the
transformer.
In the first phase of the algorithm, the principle of the percentage-biased
differential relay is used to detect internal faults at every sampling interval. The
second phase of the algorithm involves a simultaneous check on the location of
the point (idh xpk). If this point does not fall on the i^-xp curve of the transformer,
a trip signal is issued.
The basis of this approach is explained in Figure 10.8(a), which shows an
open-circuit magnetising characteristic of a transformer and the (idh tpk)
relationship for an internal fault inside the transformer. For the internal fault
condition, the terminal voltage (and hence tpk) is much smaller than is the case
under inrush conditions. Consequently there are distinct regions in the i—xp
plane that define the fault or non-fault status of the transformer.
The above technique works satisfactorily when the residual flux in the core of
the transformer is close to zero. Practically this is not always the case, and
consequently the ij-\j) characteristic of the transformer may vary as shown in
Figure 10.8(6). This situation makes it impossible to differentiate between fault
and no-fault regions in the i-ip plane if the flux linkages are computed
according to eqn. 10.33. This is because the estimated value is subjected to an
error equal to the value of the residual flux linkages.
One way of overcoming this problem is to use a restraining function which is
determined by the slope dxp/di rather than the flux ip itself. From eqn. 10.33 we
then obtain

Figure 10.9 shows that there are two regions in the dtp/di—i plane. The first
region corresponds to a fault condition or an operation in the saturated part of
the magnetisation curve. The second region, which is significantly removed
from the first, designates an operation on the magnetising curve in the
unsaturated part. It has been observed that for internal fault conditions the
current samples and corresponding flux derivative dxp/di remain continuously
in region 1. On the other hand, during inrush conditions, they alternate
between the two regions. This phenomenon can be used to create an index of
restraint kT which is increased each time a sample pair (ik, (dip/di)\k) falls in
186 Digital differential protection of transformers

region 1. However, the index is decreased whenever the sample pair enters
region 2. The index is also constrained to remain always positive. Thus
kr = kr -f l if the differential current indicates a

Figure 10.8 Transformer magnetising curve

(a) fault and non-fault region
(b) effect of remanent flux
 Digital differential protection of transformers 187

"dT

region 2: no fault region

region 1: fault region

Figure 10.9 Fault and non-fault regions in the dxp/di-i plane

trip condition and if the pair

[ih (dxp/di)\k] is in region 1
if * r > 0 , and the pair [ih (dip/di)\k]
is in region 2 (10.36)
fk-i

The value of kr is found to increase almost monotonically for fault conditions

while it never reaches greater than a threshold value krmax for all non-fault
conditions. However, the threshold £rmax is found to depend, among other
things, on the sampling rate. Therefore krmax must be determined experimen-
tally.

10.7 Basic hardware of microprocessor-based transformer

protection [17]
Figure lO.lO(a) shows the connection of a microprocessor-based protective
scheme to a three-phase, two winding A/Y transformer. The three-phase
currents at the primary and secondary sides are first reduced using current
transformers of suitable ratio, then connected to the microprocessor-based
system. Figure \0.\0(b) shows the basic interfacing arrangement to the
microprocessor system where the currents obtained from the secondary side of
the current transformers are fed to analogue amplifiers for signal amplification.
The currents are then sampled by sample-and-hold (SH) circuits and digitised
by an analogue-to-digital convertor (ADC) before being presented to the
microprocessor. The SH, multiplexer (MUX), A/D and D/A convertors are
controlled by software. At the microprocessor, the data is processed using the
188 Digital differential protection of transformers

CB CB

•m -m
-m- •m-

I—omnmr—I
A
>r'sc
trip signal trip signal;

interfacing interfacing
microprocessor system
circuits circuits

trip
control bus

f A
T
RAM DAC
ROM DAC
12 12
bit timer bit signals for

=t> /iP system

=0 process
analysis

terminal
host
computer

analogue
signal address bus
amplification

70./0 Microprocessor based protective scheme for the protection of a three-phase,

two-winding transformer
(a) connection
(b) typical interfacing arrangement
CB = circuit breaker
SH = sample and hold
MUX = multiplexer
ADC = analogue to digital converter
DAC = digital to analogue converter
 Digital differential protection of transformers 189

particular algorithm chosen. In the event of an internal fault a trip signal is sent
through the D/A convertor to trip the circuit breakers on both sides of the
transformer.
It will be noted that unlike conventional schemes, no interposing current
transformers are required for phase rotation, as this is readily implemented by
software.

10.8 References
1 KENNEDY, L.F., and HAYWORD, C D . : 'Harmonic-current restrained relays for
differential protection', AIEE 1938, 57, pp. 262-266
2 HARDER, EX., and MARTER, W.E.: 'Principles and practices of relaying in
United States', AIEE, 1948, 67, Pt II, pp. 1005-1022
3 MATTHEWS, C.A.: 'An improved transformer differential relay', AIEE, 1954, 73,
Pt III, pp. 645-650
4 ROCKEFELLER, C D . : 'Fault protection with a digital computer', IEEE Trans.
1969, PAS-88, pp. 438-464
5 EINVALL, C . H , and LINDERS, J.R.: 'A three phase differential relay for
transformer protection', IEEE Trans. 1975, PAS-94, pp. 1971-1978
6 HERMANTO, I., MURTY, Y.V.V., and RAHMAN, M.A.: 'A stand-alone digital
protective relay for power transformers', IEEE Trans. 1991, PWRD-6, pp. 85-95
7 SCHWEITZER, E . O , LARSON, R.R., and FLECHSIG, A.J.: 'An efficient inrush
current detection for digital computer relay protection of transformers', IEEE PES
Winter Meeting, 1977, Paper A7751O-1
8 LARSON, R.R., FLECHSIG, A.J., and SCHWEITZER, E.O.: 'The design and test
of a digital relay for transformer protection', IEEE Trans., 1979, PAS-98, pp. 795-
804
9 DEGENS, A.J.: 'Algorithm for a digital transformer differential protection based on
a least-squares curve fitting', IEE Proc. C, 1981, 128 (3), pp. 155-161
10 DEGENS, A.J.: 'Micro-processor-implemented digital filters for inrush current
detection', Int. J. Elec. Pow. & Energy Syst., 1982, 4, pp. 196-205
11 RAHMAN, M.A., and JEYASURYA, B.: 'A state of the art review of a transformer
protection algorithm'. IEEE Trans., 1988, PWRD-3, pp. 534-544
12 RAHMAN, M.A., and DASH, P.K.: 'Fast algorithm for digital protection of power
transformers', IEE Proc. C, 1982, 129 (2), pp. 79-85
13 RAHMAN, M.A., JEYASURYA, B., and GANGOPADHYAY, A.: 'Digital differ-
ential protection of power transformers based on Walsh functions', Trans. CEA
Engineering & Operating Div. 1985, 24, Paper 85-SP-149
14 JEYASURYA, B., and RAHMAN, M.A.: 'Application of Walsh functions for
microprocessor based transformer protection', IEEE Trans., 1985, EMC-27, pp. 221-
225
15 THORP, J.S., and PHADKE, A.G.: 'A microprocessor based three phase trans-
former differential relay', IEE Trans. 1982, PAS-94, pp. 426-432
16 PHADKE, A.G., and THORP, J.S.: 'A new computer relay flux-restrained current
differential relay for power transformer protection', IEEE Trans., 1983, PAS-102,
pp. 3624-3629
17 DEGENS, A.J., and LANGEDIJK, J.M.: 'Integral approach to the protection of
power transformers by means of a microprocessor', Elect. Pow. & Energy Syst. 1985, 7,
pp. 37-47
 Chapter 11
Digital line differential protection

11.1 Introduction

As power systems grow both in size and complexity, it becomes common to use
long and heavily loaded two-terminal lines as well as multi-terminal and tapped
lines [1-4]. This in turn has created difficult problems for their protection. Such
problems include:
(i) Distance relay underreach:
The infeed from another source to a fault point of a multi-terminal line
causes distance relays to measure a higher impedance than the actual line
impedance. This causes a delay in the tripping of the relay, which in turn
may result in incorrect sequential tripping at other terminals.
(ii) Distance relay overreach:
This problem arises when an external fault is fed from a number of
terminals. This causes overreach, in which the distance relay measures a
lower impedance than that of the fault loop. This can again cause false
tripping and difficulty in determining appropriately the time selectivity
between the protected and adjacent lines.
(iii) Effect of load current:
Load current can cause an inaccurate impedance measurement by
distance relays, which in turn can affect the integrity of the decision
reached.
(iv) High resistance faults:
A satisfactory operation of distance relays is not always achieved under
highly resistive fault conditions, e.g. an earth fault through a tree. This
type of fault is in many cases so similar to normal load conditions that the
distance relay may fail to recognise it as a fault condition.
(v) Pilot-wire limitation:
Pilot-wire relays can be applied successfully to overcome many of the
problems described above. However, this is true only if the length of the
line is not greater than typically 20 km because of the resistance and
capacitance associated with the pilot wire. Various means of pilot
compensation have been developed, but in general pilot-wire-based protec-
tion cannot be applied satisfactorily to many transmission circuits.
Recent developments in digital differential protection have been aimed at
overcoming these limitations. These schemes can be broadly classified into
current-based schemes and composite voltage- and current-based schemes.
 Digital line differential protection 191

ix(t) iY(t) Y

iF(t)
internal
fault

Figure 11.1 Basic three-terminal line configuration

11.2 Current-based differential schemes

11.2.1 Basic principles of line current differential protection [3—5]
The basic principle of line current differential protection used for two- and
multi-terminal lines is essentially the same as the percentage-differential
protection applied to transformers, which was discussed in Chapter 10.
Consider the three-terminal line shown in Figure 11.1. An instantaneous
differential current signal D(t) and a bias (or restrain) current signal B(t) are
typically formed using the instantaneous currents at the three ends such that
i,(t) + i,(*) + i<(t) (H.l)
ix(t)-h(t)-h(t) (H.2)
where ix(t), iy{t) and iz(t) are the instantaneous currents measured at ends X, Y
and Z, respectively.
Under healthy conditions, ideally, the magnitude of the differential quantity
D(t) should be zero. In practice, it has a small value both under normal
operating conditions and external faults, due to line charging current and other
errors such as a mismatch of current transformers. However, once an internal
fault occurs, the differential quantity D(t) approximates to the fault current
iF(t), i.e. the current flowing out of the fault point.

11.2.2 Frequency modulation current differential protective scheme [3, 4]

In frequency-modulation (FM) current-differential protective schemes, the
instantaneous values of the currents at each end are frequency modulated and
typically transferred to the other terminal (s) via a voice channel of a microwave
communication network, as shown in Figure 11.2. Every terminal can be
arranged to perform the functions of transmitting, receiving and initiating
tripping signals when a fault is detected, and this is commonly called a master-
192 Digital line differential protection

master-master (MMM) system (assuming a three-terminal line). The other

way of arranging the system is to install a master in one terminal only, which
receives current data from other terminals (slaves) and generates tripping
signals which are sent to the slave terminals. Such a system is called a master-
slave-slave (MSS) system.
The transmission of the instantaneous current values of each terminal is
usually achieved by allocating a frequency band, from 300 to 3400 Hz for each
current. The current at the secondary side of the CT at each terminal is
converted into a voltage, which in turn is converted into a frequency signal in
the appropriate voice band before it is transmitted to remote terminals. Either a
microwave communication link or optical fibres can be used to transmit the
necessary signals. At the receiving end, the current signal is recovered by
conversion of the frequency-modulated signal to a voltage signal.

FM relay characteristics
Although the FM differential relay relies fundamentally on the percentage
differential principle, its percentage restraining characteristics are often slightly
different from the conventional type applied to transformer protection. Figure
11.3 shows the basic characteristics of a typical percentage FM differential
relay. It will be seen that the characteristics consist of two sections, AB and BC.
The slope of section AB, the small current region, is chosen to permit the
detection of small internal fault currents under the existence of heavy load
conditions. It is also effective for detecting small internal fault currents under
high arc resistance conditions. On the other hand, the slope of section BC is
such that it can deal with heavy internal fault conditions. This section is also
effective in preventing false tripping during external fault conditions, which
cause CT saturation and thus generate significant differential-current signal
components.

A A
I/ V \
communication link

one or two terminals

may be loads
DR = differential relay
CE = communication equipment

Figure 11.2 Basic construction of FM relay system

 Digital line differential protection 193

D(t)

operating zone

block zone

B(t)

Figure 11.3 Percentage restraining characteristic of an FM current-differential relay

It is important to note that differential and bias signals are commonly derived
separately for each phase of the line. In this way, many of the problems formerly
associated with the summation of phase quantities are overcome in modern
digital line differential protection arrangements.

11.2.3 Modal current based protection scheme [5]

Operating principles and tripping criteria
The operating principles of this relay also depend on the percentage-differential
current concept. However, the differential quantity D(t) and the bias quantity
B(t) are derived using the instantaneous values of modal currents at the line
terminals. Let us first consider these principles using total time variations for a
three-terminal line. With reference to Figure 11.1, the two quantities D(t) and
B(t) at the master end are given by eqns. 11.1 and 11.2, respectively. It will be
noted that, under temporary two-terminal operation involving transmission
between ends Y and Z, or conditions where there is a complete loss of infeed at
end X, the differential signal D(t) and the bias signal B(t) will be equal in
magnitude. This is of no consequence for internal faults, but with external faults
causing an abnormally high degree of CT saturation at ends Y and Z, it could
cause sufficient unbiased differential current to be processed and thereby lead to
relay instability. This problem can be overcome by adding a logic check within
the tripping decision process, as will be explained later.
194 Digital line differential protection

In its simplest form, the tripping criterion adopted by the relay is that the
tripping signal is initiated when the magnitude of the differential quantity D(t)
exceeds that of the bias quantity B(t) by a certain predefined threshold value Ks:
\D(t)\-KB\B{t)\*Ks (11.3)
However, to avoid the above-mentioned problem, a modified process des-
cribed by eqn. 11.4 is used. In this case the differential current component D(t)
is checked against a time-variant threshold signal S(t):
\D(t)\*S(t) = Ks + KB\B(t)\ (11.4)
The basic arrangement typically used is similar to that shown in Figure 11.2 for
an MMM scheme, although MSS schemes are common when using this
approach.

Modal components of differential and bias quantities

As previously discussed in Chapter 8, assuming an ideally transposed line,
phase quantities can be transformed into what are referred to as modal
quantities. This has the advantage of producing relay measurements that do not
require complex transformations, and therefore do not delay signal components
in their derivation. The first modal component of differential and bias
quantities Dx{t) and Bx{t) is formed as the difference of the 'a' and V phase
currents, such that
DM = [»„(<) - ia(t)] + [«„(<) - i9(t)] + lUt) - iK(t)]
* i(0 = [ijt) - Ut)] - [i*(t) ~ U ' ) ] " [*«(<) - i«t)]
The second modal component based signals D2(t) and B2(t) art commonly
found from the difference between 'a' and ' b ' phase currents:
Dt(t) = [L(t) - ibx(t)] + [iv(t) - iby{t)] + [i«(t) - ik(t)]
B2(t) = [L(t) - U O ] - [iv(t) - iby{t)} - [««(<) - hM
where iax, ih, icx are the V , s b', V phase currents at end X, iay, ibp i^ are the phase
currents at end Y, and iaz, iba icz are the corresponding currents at end Z.
Since in practice there are no fault conditions that can simultaneously give
rise to zero-valued modal current signals, tripping is initiated for internal faults
by at least one modal channel. This form of signal compression reduces the
communication channel requirement from three (for separate phase-by-phase
comparison) to two.

Trip-decision logic process

A typical trip-decision logic process can best be explained by reference to Figure
11.4, where (a) shows a flow diagram of the decision process algorithm. Figure
11 A(b) shows typical differential-current waveforms for external and internal
fault conditions. The former is small and comprises mainly spill current, which
is predominantly high frequency, whereas the latter is large and predominantly
power frequency. To simplify the explanation, the pick-up level is taken as ± Ks
 Digital line differential protection 195

instead of the time variant signal ±£(/) (see eqn. 11.4), i.e. KB is assumed to be
zero.
Let us first consider the external fault waveform. It can be seen that the
magnitudes of the four samples 2-5 are above the pick-up levels ± Ks. Thus,

internal fault

Figure 11.4 Typical extended digital decision logic process [5J

(a) decision logic algorithm
(b) typical waveforms
196 Digital line differential protection

after the completion of the first operation of the logic, the decision counter KD is
set for an up count. However, since the polarities of these samples alternate
between positive and negative, as shown on the waveform, the second operation
of the logic results in a down count. Consequently KDy for the waveform under
consideration, stays close to zero at all times.
Consider now the internal fault waveform. It can be seen that once the
differential current has exceeded the pick-up level, it stays above that level for
an appreciable time. This means the two operations of the logic related to
comparisons of magnitude and polarities of four successive samples at a time
will allow KD to attain the required value of 4 very rapidly. For this particular
waveform, it can be seen that samples 2—8 successively indicate an up count.
The criteria of adopting a four sample check and a trip signal initiation at a
decision counter output of 4 is based on extensive studies aimed at maximising
relay stability under external fault conditions and at the same time maximising
the sensitivity to internal faults. The adoption of the bias quantity KB\B(t)\ (see
eqn. 11.4) further enhances stability under healthy conditions without signifi-
cantly affecting performance for internal faults. In essence, the trip decision
logic described makes use of further features of the relaying signals to enhance
the integrity of the decision asserted under onerous conditions, which may
otherwise cause relay malfunction.

11.3 Composite voltage- and current-based scheme [6, 7]

Although current differential schemes are relatively fast in terms of fault-
clearance times, they have the common problem of becoming insensitive to low
levels of fault current because of their total dependence on current. This arises
from the standard practice of providing a bias signal proportional to through
current to ensure security under conditions where a significant differential
signal is generated under healthy conditions, e.g. due to shunt capacitance
charging currents or current transformer errors. A second problem is that
certain system operating conditions may result in the difference between the
differential and restraint signals being small for internal faults, which means a
reduction in dependability of the protection.
To overcome these problems, an approach has recently been adopted which
involves deriving differential signals that are functions of both the voltages and
currents measured at each end of either plain or teed lines. Consequently, a bias
signal to compensate for capacitance spill currents is no longer required. This in
turn increases the relay sensitivity to a much lower level of fault currents than
provided by schemes that are totally dependent on current measurement.
11.3.1 Basic operating principles
The basic operating principle relies on deriving signals proportional to the
instantaneous values of modal voltages and currents at the three ends of the
line. Ideally the sum of these signals should be zero under healthy conditions
 Digital line differential protection 197

e
|y Y

Figure 11.5 Basic two-terminal line configuration giving current-reference directions

but significantly finite under internal fault conditions. Thus, for a three-
terminal line such as that shown in Figure 11.1 we have

{ = 0 healthy conditions
,r , J-.
# 0 internal fault conditions
* = 0> !>2 (11.7)
v
'
For a two-terminal line the signal Sz is equal to zero and the above equation
reduces to

{ = 0 healthy conditions
^ O i rinternal, fault
c..u -conditions
__^:._. * = 0,l,2 (11.8)

where Sx, SY and Sz are the terminal signals at ends X, Y and Z, and k is the
mode number.
11.3.2 Formation of terminal signals
113.2.1 Two-terminal lines [6]
Assuming a distributed-parameter, transposed three-phase transmission line,
the modal voltage and current relationships in the frequency domain can be
described by eqn. 8.30:
2
), - ^ = ( y » ) ^ * = 0, 1, 2 (11.9)

As previously explained in Chapter 8, the above equation shows that wave

propagation in a three-phase line can thus be considered in terms of three
independent components, each possessing its own modal propagation constant
y^)(co) and associated modal surge impedance Z^((w), where k = 0, 1 and 2.
The solutions of the voltage and current differential equations shown in eqn.
11.9 take the form
V™ ((*)) = K{ fry(k)x + K2 e yW * (11.10)
and

where K{ and K2 are arbitrary constants and x is an arbitrary length of line.

In the above equation, the frequency variable a) has been dropped from the
propagation constant y{k) in order to simplify the notation.
The arbitrary constants Kx and K2 can be determined from a knowledge of the
boundary conditions for a particular system. Consider the two terminal line
shown in Figure 11.5, which has the following boundary conditions:
at end X
at end Y <1M1>
198 Digital line differential protection

where / is the length of the line.

By substituting eqn. 11.11 into eqn. 11.10, we obtain the voltage and current
at end X:
) (1 2)
k = 0, 1,2
) = (K]-K2)/ZP(a>)
and those relating to end Y will be

By substituting the values of Kx and K2 obtained from eqns. 11.12 into eqns.
11.13, we finally obtain
^(aO+ZjPM/pV^e^ * = 0, 1, 2 (11.14)
By using the current and voltage reference directions shown in Figure 11.5, i.e.
by reversing the direction of the current IY at end Y, eqn. 11.14 becomes
n\(^)-Z0{(D)lf((o) = ^l[nK(o) + Z^\(o)lf((D)^ * = 0, 1,2 (11.15)
or S f + 5^ = 0 * = 0, 1,2 (11.16)
where Sx{k) is equal to the right-hand side of eqn. 11.15 and S$ is equal to the
left-hand side negated.
113.2.2 Three-terminal lines
The basic theory laid down above for the two-terminal lines can be easily
extended to include teed (or three-terminal) feeders. Consider again Figure
11.1, which shows that a three-terminal line basically consists of three separate
line sections. These are sections XT, YT and ZT. For each of these sections an
equation similar to eqn. 11.14 can be derived in terms of the current and voltage
at the line ends. When the resulting equations are combined together (and
applying the fact that the sum of currents at the tee point T is zero), we obtain
Sf + Sf + S$ = 0 * = 0, 1,2 (H.17)
where
Sf = AxV<£\a>) + BxZW(a>)l$(fi>)
Sf = [AY- Br) Vf(a>) + (Ar+ BY)Z^{(o)I^{w)
Sf = (Az-Bz) Vf(o,) + (Az

Ay=

hi h-> h. a r e the lengths of sections XT, YT and ZT respectively.

11.3.2.3 Trip-decision logic
It is seen from the foregoing that, in theory, the signals at the ends of a line
(whether of the two- or three-terminals type) should sum to zero under all
 Digital line differential protection 199

healthy conditions. However, because of quantisation, transducer errors etc, it

becomes necessary to apply a small threshold Th to the sum signal of eqn 11.17.
Thus in practice the simplest relay tripping decision would be
\S{£\+Sf + Sf\>Th k = 0, 1,2 (11.18)
The above equation applies to a three-terminal application. In the case of a
two-terminal line S$ becomes zero.
It will be noted that the tripping decision described by eqn. 11.18 is fairly
simple compared with that used in the previously described current differential
schemes. In consequence, a relatively simple trip-decision logic process can be
used to provide the necessary degree of dependability and sensitivity. The
simplest criterion that provides an adequate performance for most protective
applications involves the initiation of a trip signal if two consecutive samples of
the signal summation of eqn. 11.18 exceed the fixed threshold setting Th.

11,4 References
1 AIEE Working Group of the Line Relay Protection Sub-Committee: 'Protection of
multiterminal and tapped lines', AIEE Fall Meeting, October 1960, Paper CP60-1274
2 'IEEE Study Committee report on protection aspects of multi-terminal lines', IEEE
Report 79, THOO56-2-PWR, 1979
3 KITAGAWA, M. et al.: 'Newly developed FM current-differential carrier relaying
system and its field experience', IEEE PES Winter Meeting, 1978, Paper F78, 291-
297
4 YAMAURA, M., MASUI, M., and OKITA, Y.: 'FM current differential carrier
relaying', Developments in power system protection, (IEE Conf. Publ. 185, 1980,
pp. 156-160)
5 AGGARWAL, R.K., and JOHNS, A.T.: 'The development of a new high speed three-
terminal line protection scheme', IEEE/PES Summer Meeting, 1985, Paper 85 SM
3200-3207
6 AGGARWAL, R.K., and JOHNS, A.T.: 'A differential line protection scheme for
power systems based on composite voltage and current measurements', IEEE/PES
Winter Meeting, 1989, Paper 89WM 053-0 PWRD
7 AGGARWAL, R.K., and JOHNS, A.T.: 'New approach to teed feeder protection
using composite current and voltage signal comparison', Developments in power
system protection (IEE Conf. Publ. 302, 1989, pp. 125-129)
 Index

Accuracy 13, 55, 89, 90, 107 Conversion subsystems 44—56

Admittance 126 Convolution 46, 175, 178
Aerial modes 126, 135 Cost/benefit ratio 3
Algorithms 4, 57, 60-69, 72-114, 146, 174- Counter controlled converter 52, 53
187 Current differential protection 171-189, 191-
differential equation based 103-114 196
finite-duration impulse response filter based flux-restrained 183-187
174-179 frequency modulation (FM) 191-193
flux-restrained current differential 183-187 modal current based 193-196
Fourier-analysis-based 72-84, 181-183 Current transformer (CT) 39, 41, 171, 172,
fractional-cycle window 75-79 187, 192, 193
full-cycle window 73-75 Curve fitting 18-22
least-squares-based 92-102, 179-181 Cutoff frequency 42, 43, 46
sinusoidal-wave-based 60—69
travelling-wave-based 146 D'Alembert approach 168
Walsh-function-based 84-91 Data window 1, 72, 76-81, 83
Aliasing 46, 47 DC offset 64, 65, 67, 139
Aliasing error 46 reduction 65, 67
Analogue filtering 41—43 Dependability 154, 159, 196, 199
Analogue interface 152 Difference function operators 10-12
Analogue multiplexer 40 Difference functions 7—10, 62
Analogue-to-digital converter (ADC) 40, 52— backward 8-10, 62
56, 58, 187, 188 central 9, 10
counter-controlled 52, 53 forward 8-10
dual-slope 53-55 Differential equation algorithm 2, 4, 103-114
parallel-comparator 55, 56 with selected limits 107-110
Anti-aliasing filters 42 Differential protection 60, 171-189, 190-199
Average operator 11,12 Differential relays 172-174, 183-187, 191-199
biased 172, 173
Backward differences 8-10 composite voltage-and-current-based 196-
Backward interpolation 14 199
Backward waves 121 flux-restrained current 183-187
Bergeron's-equation based protection 148—152 frequency modulation (FM) 191-193
Bergeron's equations 140—142 harmonic-restrained 172-174
single-phase lines 140, 141 modal current based 193-196
three-phase lines 142 percentage-biased 172, 185
Biased differential relaying 172, 173 Differentiation 16-18
Binary numbers 31, 33, 51, 52 Digital filtering 56, 57, 107-110, 181, 182
Busbar protection 4 Digital multiplexing 49, 50
Digital relay subsystem 56-58
Cable circuit protection 4 Digital smoothing 65
Capacitor voltage transformer (CVT) 41 Digital-to-analogue converter (DAC) 50, 51,
Central difference interpolation 15 187-189
Central differences 9, 10 Directional comparison relays 2
Clark transformation 125, 164 Directional discrimination 164, 165
Comparators 52—56 Discriminant function based protection
Composite voltage-and-current-based schemes 164, 165
differential protection scheme 196-199 Discriminant functions 143—146
 Index 201

Distance relays 1, 2, 61, 69, 70, 190 Impedance 1, 60-62, 66-69, 72, 79, 101, 102,
Distribution systems 79 126
Dual-slope converter 53—55 variation with time 7
Impulse responses 174-176, 178
Earth mode 126, 135 Inrush detection algorithm 180, 181
EHV applications 152 Integral least squares fit 92-95
Elliptical trajectory based technique 136, 138- Integrators 53, 54
140, 166-168 Internal fault detection 148-170
Errors 13, 46, 56, 62-65, 76, 79, 92, 93, 98, Interpolation formulas 12-15
185 backward 14
central difference 15
forward 12, 13
Fault direction 153, 154, 164-170
Gregory-Newton 13, 14
Fault discrimination 178-181
Laplace-Everett 15
Fault location 4, 61, 127-140, 143-146
Newton-Gauss 15
forward faults 131-135, 144
Stirling 15
reverse faults 129, 130, 132, 134, 135, 145
Fault trajectory 139, 140
Kalman filters 1, 2
Finite difference techniques 7-12
Karrenbauer transformation 125, 164, 165
Finite-duration impulse response (FIR) filter
174-179
Least squares methods 4, 18-22, 92-102,
extraction of fundamental and second-
179-181
harmonic components 175-178
integral 92-95
fault discrimination 178, 179
multi-variable series 99-101
Finite-duration impulse response (FIR) filter
power series 95-99
based algorithms 174-179 LFDC digital relay 164
Flexibility 3 Logic circuits 154, 161, 194-196, 198, 199
Flipflops 56 Lossless single-phase lines 119-123, 140, 141,
Flow charts 56, 57, 164, 165 143-146
Flux-restrained current differential relaying Low-pass filters 40-43, 45, 46, 110
183-187
Forward differences 8-10 Master-master-master (MMM) system 191,
Forward interpolation 12, 13 192, 194
Forward waves 121 Master-slave-slave (MSS) system 192, 194
Fourier analysis 24—30 Memory 3, 40, 57, 70
Fourier analysis based algorithm 72-84, 181 — Microprocessors 187
183 Microwave communication links 192
Fourier-transform based 79-84 Modal analysis 123-126, 135, 142, 146
full-cycle window 73-75 Modal components 124, 194, 196, 197
half-cycle window 75-77 Modal current based protection scheme 193-
sub-cycle window 76-79
196
Fourier coefficients 26, 27, 34-37
Modal equivalent circuits 123
Fourier kernels 84
Modal propagation constant 125, 126, 197
Fourier series 1, 4, 24-27, 72, 73
Modal surge impedance 125, 126, 197
complex form 25
Modal transformation matrices
trigonometric form 24, 25
Clark 125
Fourier transform 1, 28-30, 79-84
Karrenbauer 125
Fourier-transform based algorithm 79—84
Wedepohl 125
Fourier-Walsh transformation matrix 37, 85
Multiplexers 40, 43, 44, 49, 50, 58, 187, 188
Fractional cycle windows 75-79
logic circuits 49, 50
Frequency domain 46, 72, 197
Multi-terminal lines 190-199
Frequency modulation (FM) relays 191-193
Multi-variable series least squares technique
Frequency responses 75-77, 175, 177, 182
99-101
Generator protection 171
Noise 2, 65, 67, 160, 161, 164
Numerical differentiation 16-18
Haar function 183
Nyquist frequency 44
Harmonic-restrained differential relaying 172—
174 Optical fibres 192
Harmonics 92, 99, 100, 107-110, 172-183 Out-of-step blocking 70
202 Index

Overcurrent protection 70 Surge impedance 120-122, 125, 129, 135

Overcurrent relays 3 Surge protective circuit 40-42
Overreach 153, 190 Surge transit time 141
Switchgear protection 4
Parallel-comparator converter 55, 56
Performance 3 Teed power circuits 150, 191, 192, 198
Periodic functions 25—27 Three-phase transposed lines 123-125
Phase angle 68, 88-90 Time domain 46
Phasors 73, 76, 104, 124 Transducers 39-41
Pilot-wire relays 190 Transformer protection 95, 171-189
Plain feeders 150 Transformers 172, 183-189
Post-fault analysis 4 magnetising characteristic 185, 186
Power series least squares fit 95-99 two-winding 184, 187, 188
Prodar 70 69, 70 Transients 115
Propagation constant 119, 125 Transmission line algorithms 60, 95, 103—114
Propagation velocity 120, 129 lumped series impedance based 110-112
Pseudoinverse approach 21, 22, 101 single PI section based 112
Transmission line protection 103, 107-110,
RALDA 152-154 148-170, 190-199
Random-access emory (RAM) 188 Transmission lines 2, 103-107, 110-146
Reactance 69, 115, 139 fault types 105-107
Read-only memory (ROM) 188 inductance 103-107, 110-114
Real time 62 models 2, 103, 110-112, 115-119
Rectangular transforms 183 resistance 103-107, 110-114
Relaying signals 134-137 single-phase 119-123, 136-138, 140, 141,
Relaying time 154 143-146
Reliability 2, 3, 67 three-phase 103-105, 116, 123-126, 135,
Replica surge impedance 134, 135 142, 146
U H V 2 , 3, 115
Sample-and-hold (SH) circuits 40, 47, 48, 58, Trapezoidal rule 81, 83, 86, 88, 94, 184
187, 188 Travelling wave relay 2
Sampling 44-47 Travelling wave schemes 4, 148-170
Sampling frequency 44-46, 75 Bergeron's equation based 148-152
Sampling function 44, 45 discriminant function based 164, 165
Sampling theorem 44 superimposed component trajectory based
Scratch pad memory (SPM) 70 166-170
Second harmonic ratio 181 ultra-high-speed polarity comparison 152-
Security 159, 160, 196 154
Sensitivity 196, 199 ultra-high-speed wave differential 154—164
Shift operator 10-12 Travelling waves 82, 115-146
Shifted time reference frame 95, 96 backward 121
Shunt capacitance 115 coefficient of reflection 121 -123
Signal conditioning subsystems 39-44 forward 121
Simultaneous differential equation techniques propagation constant 119
110-114 propagation velocity 120
lumped series impedance based 110-112 Trip signals 150, 152, 154, 159, 172, 174, 188,
single PI section transmission line based 189, 191-196, 199
112 Truth tables 164, 165
Sinusoidal-wave based algorithm 60—69
first-and-second-derivative method 65—67 UHV applications 152
sample and first-derivative method 60—65 Ultra-high-speed polarity comparison scheme
three-sample technique 68, 69 152-154
two-sample technique 67, 68 Ultra-high-speed relay 152-164
Smoothing 22-24, 65, 67 Under-reach 153, 154, 190
Stability 172, 193, 196
Superimposed-component-trajectory based Voltage transformer (VT) 39, 41
protection schemes 166—170
Superimposed quantities 126-139, 143-146 Walsh coefficients 34-37, 85, 86, 88-90
elliptical trajectories 136, 138-140 Walsh-function analysis 30—34
polarity 127, 128, 136 Walsh-function based algorithm 84-91
 Index 203

Walsh functions 1, 4, 30-34, 183 digital 159-164

discrete 31-34 Wave-differential relays 155-164
index 31 Wave-differential schemes 154-164
multiplication 31 Waveforms 1, 26-30, 60, 61, 64, 65, 72, 79,
symmetry 31 80, 92, 95, 100, 115, 121, 179, 194-196
timescale 31 Wedepohl transformation 125, 164
Walsh series 34, 84-91
Wave-differential protection 154-164 Zener diodes 41
analogue 155-159
DIGITAL PROTECTION FOR POWER SYSTEMS
Digital protection is based on the use of computers
in power line relaying. Since the late 1960s, digital
devices and techniques have been applied to almost
all new protection schemes. Today the technology is
moving towards standardised hardware platforms; at
the software level, however, there remains a huge
variety in approaches and protection algorithms.
This book gives a fairly detailed understanding
of the principles and techniques underlying the
application of digital technology and algorithms to
protection. It avoids going into detail of specific
products: up-to-date information on these is available
from the manufacturers. Instead it aims to give the
reader a thorough understanding of the generic
problems of digital protection.
The text covers the mathematical basis of
numerical techniques and relay algorithms, the basic
elements of digital protection and the fundamentals
underlying the commonest algorithmic forms,
particularly as applied to line protection. It deals with
the fundamentals of travelling-wave techniques and
their application to transmission lines, and with
digital differential protection of transformers and
lines.

Allan T. Johns is Professor of Electrical Engineering

and Head of the School of Electronic & Electrical
Engineering at the University of Bath. He is the author
of over 200 journal papers, and in 1982 he was
awarded the degree of Doctor of Science for his
original and substantial contribution to knowledge in
the area of digital protection and simulation of
electrical power systems.
He worked for nine years in British industry, both in
electricity supply and in manufacturing, and is still
heavily involved in industrial consultancy. He is a
Fellow of the IEE, a Series Editor for the IEE Power
Series of books and Tutorial Editor of Power
Engineering Journal. Among the awards he has
earned are the IEE Crompton Premium in 1983 and
the IEE Power Divisional Board Premium in 1989.

Salman K. Salman is with the School of Electronic

& Electrical Engineering at the Robert Gordon
University in Aberdeen. His research interest covers
analysis and protection of power systems. A graduate ISBN 978-0-86341-303-X
of Baghdad University, Sheffield University and
UMIST, he is the author or co-author of more than 35
papers. He spent six years working in the electricity
supply industry, and before his current position was
Associate Professor at the Behghazi University,
Libya. Dr. Salman is a Member of the IEE and a
Senior member of the IEEE.
Peter Peregrinus Ltd.
The Institution of Electrical Engineers,
Michael Faraday House,
Six Hills Way, Stevenage,
Herts. SG1 2AY, United Kingdom
Cover photograph reproduced by permission of ISBN 0 86341 303 X
GEC Alsthom T & D Protection and Control Ltd. Printed in the United Kingdom