Combined wavelet transform and regression

technique for secondary current compensation of

current transformers
F. Li, Y. Li and R.K. Aggarwal

Abstract: A technique is proposed to effectively and efficiently compenyate for secondary current
of current transformers. particularly in the presence of remanent flux. The technique first utilises
the wavelet transforni to accurately distinguish the distorted section from the healthy section of a
secondary current wavefonn. Secondly the technique uses the I-egression method to recover the
distorted section making use of features extracted from the healthy section. essentially to restore the
secondary current to a scaled down replica of the primary system current. The proposed technique
is demonstrated on a samplu power system, achieving an average compensation error of I.5%
under varying degrees of saturation caused by different levels of reinanent flux.

1 Introduction which arises by virtue of the fact that when a circuit breaker
interrupts a fault. remanent flux gets trapped in the CT
Protective relays play an increasingly important role in cver- core. which could last almost indefinitely under load
stressed electric power systems. For relays providing conditions. If the impedance of the CT load has a phase
inaxinium protections in undesirable conditions such angle of 60" lag, the remanent flux would reach about 50%
as faults, an accurate knowledge of current signals is of the peak flux value. When a second fault of the same
essential. A practical way of obtaining this knowledge is polarity occurs, the transient flux rises l'rom the already
to use current transformers (CTs) to step down high trapped remanent flux. As a result, the effective excursion of
currcnts in the primary windings to lower levels in flux before saturation ensues will be much less. Hence,
secondary windings both in wavefonn shapes and inagni- saturation will take place earlier and will be much more
tude [I]. CTs are ~ioiinally used to feed information severe. Tests carried out by a CalladIan company have
into measuring instruments and protective equipment revealed that there are many conditions under which
and tils0 to isolate the equipment from the primary circuits remanent flux is left in a CT core. The tests with CTs on
to ensure the safety of equipment and personnel. Ideally. a 230 kV system have shown that 27% cases had between
the secondary current of a CT should he of the 60 and 80% of peak flux density remaining in the core when
same Waveshape as its corresponding primary current the primary circuit breaker was open [4]. Fig. 1 illustrates
with ii scaled-down magnitude for correct relay operations. the significant effect of remanence on the secondary
CTs widely used in power system consist of thi-ee currents.
main parts, namcly. a core of ferrous material, a priinary
winding and a secondary winding. The core is magnetised
by current in the primary winding. From a modelling
point of view, this magnetising efkcct is represented
by a magnetising branch in the equivalent circuit of the
CT. The current flowing through the branch is the exciting
current of the CT: this is one of the major causes of the
measurement error, thereby influennng the performance of I
a CT [ 2 4
A principal cause oFCT error is due to saturation of the
iron core, i.e. inability of the latter to support a large flux
swing emanating from a large DC offset under transient
conditions. Thc saturation effect in a CT will lead to the
secondary current to be distorted under transient condi-
tions. Very often the unwanted distortior can be accen-
tuated with thc presence of renianent flux in the CT core
-5 t
IEE. Z W
I/% P r o m d i n g . ~online no. 20020296
Dol. l @ . l i l 4 9 / i p - p ~ d : 2 i ~ i ~ 2 ~ ~ 9 ~
Publicxion d a c : 27rh May 2002. Paper 6ni recavud 13th February 2Dol aiid in
revised Corm ?Xi11 N o r e m k 2001
The authors itre with the Department of Elstrunic Eleclfical Engineering,
Uniievsiry oiBsth. Bath RAZ 7AY. VK
 The general experience with conventional static or The fault current reaches its limit when a three-phase
electromagnetic relays does not indicate that remanence is fault occurs on the transmission line close to the CT end:
a serious problem [2], but it can cause serious problems with' which is approximately 5kA RMS (Le. 300kV/60Q).
digital relays. Distorted secondary currents can threaten Although an extensive series of studies were performed to
dependability and security of relays, especially current evaluate the performance of the compensation techniques,
differential relays; this influence can have a detrimental the results presented here restrict to single phase-ground
effect on system operation. Much effort has been put into and three phase faults lOkm away from the CT location.
the development of effective and efficient compensation The EMTP (electromagnetic transients program) package
methods for CT saturation. Papers [S, 61 present methods of was used to simulate the faults with a sampling rate of 200
using auxiliary electric equipment for compensation. The samples per cycle.
equipment has shown satisfactoly compensation results
under normal Steady-state condition, hut its performance
has as yet to be validated for fault conditions. In papers 2.3 CT characteristics
[7, 8]_an artificial neural network technique is discussed. Two models are available within the EMTP that can
However, it can only partially recover distorted currents, represent CTs [IO]. One is the saturable transformer
and does not take account of remanent flux trapped in a CT component. The magnetising branch is represented intem-
core. Another algorithm based on solving differential ally as a nonlinear inductor, whereas 1.A characteristics can
equations at discrete time steps has been reported in 191. he derived using supporting program SATURATION. The
Here again, the effect of remanent flux is not considered. other one is the pseudononlinear reactor component (type
In this paper, a technique is proposed to compensate for 96). To use this model, CT hysteresis characteristics should
distorted CT secondary currents caused by CT saturation he evaluated in advance. A support routine. HYSTER-
under transient faults, particularly in the presence of varying ESIS, exists within the EMTP for calculating the character-
degrees of remanent flux in the CT core. The proposed istics. Although both models give satisfactory results for
technique first uses wavelet analysis to detect the onset of most simulation cases [I I],the latter was employed for its
distortion due to CT saturation, and a regression method is ease of inittalising remanent flux in CT core using type-96
then utilised to recover the distorted section using the inductor. The 4005 (a turns ratio of 80) tap was selected
samples from the healthy section of the waveform. The test from a 6005 ClO0 CT. Its hysteresis characteristic is shown
results attained from the sample power system under a in Fig. 3.
variety of different system and fault conditions clearly
demonstrate satisfactory compensation achieved by the
proposed technique.

2 Power system configuration 0.4 t

saiuialion Doinl
(9.52, 0.48)
The sample power system considered comprises of two
sources and a single line shown in Fig. 2 to demonstrate the
effectiveness of the proposed compensation technique for
CT saturation. GI and G2 are equivalent sources, with Z,, -0.2
and Zsz as their equivalent impedances, respectively. The
details of the power system configuration follow. -0.4
I
0 5 10 15
currem. A

Fig. 3 Mngneti.sution cww of selected CT fap

2.4 CTburden
Under maximum fault current, the secondary current
CT (neglecting the magnetising current) becomes 62.5 A
Fig. 2 Confrgurafionof the sample power s y e m (i.e.S000/80). When selecting the 400:s tap, the ClO0 CT
can supply 100*(400/600)= 66.67 V. Therefore the max-
imum load on the CT secondary is 1.10 (= 66.7 V/62.5A).
2.7 Sources Because of the existence of lead reactance and secondary
reactance, the reactive component of the CT load is h e d as
phase-ground voltage: 300 kV RMS 0.50. Hence the CT burden used is (O.S+JU.5) Q.
frequency: 60 Hz
GI positive sequence impedance: 7 +j60Q 3 Principles of proposed compensation technique
GI zero sequence impedance: I4+jl20Q The compensation technique presented comprises two
G2 positive sequence impedance: 0.45 +j4.5 R stages; the first involves accurately distinguishing between
G2 zero sequence impedance: 0.45 +j4.5Q the healthy part and the distortion caused by CT saturation
in the current waveform, and the second entails the
employment of a regression method to mitigate the effect
2.2 Transmission Line of distortion in the current waveform through compensa-
tion. In this respect, it is important to note that any
length 161 km distortion in the current signals other than that caused by
R1=0.0306Q/km. x1=0.365Q/km, C1 = 12.03nF/km CT saturation is left intact by the proposed compensation
technique, and is simply reproduced as a scaled-down
RO=0.304Q/km, XI= 1.358R/!un, CO=7.52nF/km version of the input waveforms into the CT primary.
498 IEE Proc-Gene?, Tramri Dirtrib., Vol. 149, No. 4. July 2002
3.1 Introduction to wavelet analysis that the resultant coefficients reveal high-frequency features.
The Fourier transform (FT) is a well-known and powerful In contrast; a high scale means a stretched wavelet and a
tool for signal spectral decomposition. However, it has a large window size, and therefore the coeixcients reveal low
major drawback of not being able to indicate when sudden frequency features. A coefficient obtained from steps (iii) to
changes take place. A solution to this problem is the timc- (v) represents how closely correlated the wavelet is
dependent Fourier transform, also known as the short-time compared with the section of the signal under study.
Fourier transform (STFT). STFT processes data in a small The most commonly used WT method is the discrete
section of a signal at a time. which can be looked upon as wavelet transform (DWlJ with which the number of
an invisible window containing a finite number of sample decomposition levels can he easily adjusted to meet various
points being analysed. This window moves along the signal practical requirements. Fig. 5 is an example of a one-level
as a continuous process until the whole signal is covered. decomposition implemented by DWT, in which the source
Once the window size is selected with STFT, it remains signal is a truncated sine wave and hence its first derivative
fixed throughout the signal processing. This approach does, (FD) is discontinuous, The first level approximation and the
however, affect the accuracy of the information if the first level detail decomposed from the source signal are also
window size is not appropriate. The wavelet transform shown in the same Figure.
(WT) overcomes the fixed-window problem by allowing
variable-sized windows to be utilised, and gives detailed
frequency as well as time-domain information, Le. precisely
when a sudden change occurred [12].
A wavelet is a waveform having limited duration with an
average value of zero. A typical Muses several versions of
a chosen mother wavelet to give different levels of frequency
details. These versions are different in their time durations
(known as scales in wavelet). Applying WT to a signal is
essentially to reveal the extent of similarities between the
selected wavelet and the windowed sections in the signal,
ultimately to expose different features of the signal, such as
abrupt changes, higher harmonics or the trend of the signal.
The following is a simplified procedure of WT implementa- detail approximation
tion:
Fig. 5 Di~corrtitiuily.shown by firs1 Iecel d a d of d x r e l e ~ r u ~ d e r
(i) Select a wavelet. tram/onii
(ii) Select the section at the start of the signal (as if the
section is put in a window), and compute a coefficient The source signal is aligned intentionally to the fist detail
corresponding to the section. in Fig. 5 to demonstrate that the first-level detail precisely
detects the inception and completion points of FD
(iii) Shift the wavelet to the right of the signal (as if the discontinuity through large magnitudes of resultant coeff-
window is moving), and repeat step (ii) until the whole cients. T h s attribute of the DWT has had important
signal is covered. The combined effect of steps (ii) and (iia) is applications in power systems: typical examples are
technically expressed as ‘perform a one-level wavelet monitoring voltage sags and flickers (131 and studying
decomposition of a signal’. faulted networks [14]. In the study present, the DWT is used
(vi) Scale (stretch) the wavelet (as if the window size is primarily for detecting FD discontinuity, that is, to dctect
enlarged), and repeat steps (ii) and (iii). the inception and completion of the distortion of a
Repeat steps (i) to (iv) for all the desired scales, or, all the waveform. The DB4 mother wavelet has been chosen in
desired levels. this study as it best serves the purpose of distortion
detection (due to CT saturation) among other mother
wavelets available; this is owing to its ability in detecting low
The above procedure is illustrated in Fig. 4. A low scale amplitude. short duration, fast decaying and oscillating type
means a compressed wavelet and a small window size. so of signals, typical of those associated with CT saturation.

3.2 Detection of distorted current waveforms

using wavelet transform
Signal A fault current flowing through the primary circuit o r a CT
installed on a transmission line can be considered as the
wavelet
superimposition of a sine component and an exponential
component, given as
a
i(t)=Ce-”‘+cqsina,=i,,(t)+i,,(f) (1)

where i = fault current, I = time, ieXp

= exponential compo-
signal
nent, irin= sine component, E = decaying coefficient of
WaYelet exponential component, 4, =phase angle at time instant I ,
I----
L
I -
I
C= initial value of exponential component, and
C, = amplitude of sine component.
If the CT is not saturated, it produces a secondary
current of the same shape but with a scaled down value at
each sampling point, and the FD of the current signal
should be continuous. The first-level detail obtained from
4Y9
 DWT decomposition of the secondary current should he alternative form:
similar to the waveform shown in Fig. 6u.
&,.(t) = C4 sin(2n.fr + 4) (5)
where /=power frequency and cb = initial phase angle of
sine component.
There are now two more unknown parameters C, and $
to be determined. For simplicity. ( 2 n f t t 4 ) is replaced by a
single variable $ p It is difficult to have one technique to find
all the six parameters. Thus an alternative method is sought
for determining &
Like (4), variable f in is,"([) is substituted hy the number
of sampling points n. Thus the sine component becomes:

secondary isi,,(n) = C4 sin $,, (6)

current
where 4pl= phase angle of nth sampling point. Convention-
a b ally, if the phase angle or the initial point is determined as
Fig. 6 Di.storrurl purl of second[iry current retided byJir.si l e d a reference, the phase angles of the successive points can
detail be easily obtained. However, in real-time, it is difficult
(I Secondary current not distorted to determine the initial phase angle of a fault current.
h Secondary current distorted Therefore another sampling point has to be selected to
act as the reference. The reference is chosen as the first
tuming point (irrespective of whether it is peak or trough)
If the CT is Saturated, the secondaly current will he after the distortion has been deduced, which is shown in
distorted. The first-level detail from its DWT will be Fig. 7. The phase angle of this point is assumed as 742;
discontinuous, as shown in Fig. 6h. The short bursts that however, it could be - 4 2 . If the turning point is a trough
contain large coefficients of the first level detail, explicitly value, the sign of C, and I in (6) is negative; otherwise, it is
mark the start and end of the current distortion, hence positive. nni2 denotes its sampling number in Fig. 7.
clearly distinguishing the distorted section from the healthy
section of the current waveform.
prim. current
3.3 Regression technique
Once the distorted scction is separated from the healthy I.
-,.
section of a waveform it is possible to use a regression
technique to compensate the distorted section with features
extracted from the healthy section. Regression is a
technique to determine an approximate function by
performing a k d s l mean square fit to given data [IS]. In
this study the technique is used to find approximate
secondary current waveform functions by fitting the sample
points extracted from the healthy section.

3.3.7 Approximating the exponential compo- !

nent: The exponential component e-" in (I)can be
approximated by a Taylor polynomial. given by: d2-2n(n-n,,*)/*r d2
phase angle
Fig. 7 P h s e ungle of nth .sumpling poim
It has been ascertained through extensive studies that
the third-order Taylor polynomial gives a satisfactory
accuracy, and any higher-order terms are of secondary Suppose that the secondary current is sampled at a rate
importance. Multiplying (2) with C yields of S, samples per cycle, the sampling interval phase angle is
2n/& The phase angle & at sampling point n can be
derived from (7)

= C" + C,f+ C2f2+ c3ri (3)

(7)
where C, = C _C, = --St[, C2= Ca2/2,C3= -Ca3/6, C,,
C I - C? and C3 are unknown coefficients that need to
be determined by the regression method. In a digital 3.3.3 Compensating distorted section with
relay, data is digitised into discrete values through regression method: The superimposing of the coinpo-
sampling. The variable f for time can be substituted by nents discussed in both Sections 3.3.1 and 3.3.2 aves a
the number of sampling points n. Thus (3) becomes formula of the form:
i e X p ( n )= C" + Cln + C2n' + C p 3 (4) i ( n ) = C" + ~ l +nCp' + in' + C, sin 4n (8)
This formula approximately represents a current waveform
containing both healthy and distorted sections. In (X), 4nis
3.2.2 Approximating the sine component: The computed via (71, n is the sampling number, and C, C,, C2,
sine component Cash$, in ( I ) can be rewritten in an Ci and C, are unknown coefficients.
500 Ir;E P m c - G ~ w r 7
. i w " Disfrih.. Vul. 149, N o 4, July ?WZ
 Using the sampled points in the healthy section (the
number of points depends on the chosen sampling rate .....
within the digital relay), Co, C,, C, C, and C4can be solved
by the regression method performing the least mean square 3
fitting, thereby compensating for the distorted secondary
current. 2 0

4 Results and discussion

-5
Up to 20 cases comprising of single-phase to ground and
three-phase faults were simulated in this study. The 60
inception angles ranged from 0 to 180", and remanent flux i
from -75 to 80% of the flux at the saturation Doint .i :.
(hereafter symbol $,, denotes the CT core flux ai the 40
;e
saturation point). Detailed information of the simulated i

faults is listed in Table I.Although the study only considers :20

0
single-phase-to-ground and three-phase faults, the technique
is generic enough to apply to other types of faults, such as
0
phase-phase. 20 40 60
time, ms
Table 1: Initial conditions of simulated faults
Fig. 9
Fault type Inception Remanent flux in CT core
angle deg

Phaseground 0 VS %b,
-0.366 -75
45 -0.144 -30
9 5
0 0
Three-phase 90 0.096 20 L

2 0
180 0.29 60
0.38 80
-5
b,: CT core sattiration flux

Figs. 8-12 demonstrate some typical compensation

results using the proposed technique for five three-phase-
loo I after compensated

fault cascs. All cases have the different levels of remanent

flux in the core. The fault currents comprising the maximum
DC offset with such an inception angle will lead to the
severest distortion of the secondary currents under certain

5
9
conditions. However, the Figures clearly show that the
DWT accurately detects the distorted sections of
the original secondary current waveforms, which are clearly
marked with rectangles. The Figures also show the primary,
-5 original secondary and the compensated secondary
currents.
......... beiore comp. Fig. 8 shows the compensating result in the case of a
- aflercomp. remanence of -75% $, in the CT core. The polarity of the
remanence is opposite to that of the DC offset, which
results in a slight distortion beginning at the third cycle of
the fault secondary current, as shown in Fig. 8.
The extent of the distortion becomes gradually more
severe as the remanence increases rrom a negative value to a
positive value. The case shown in Fig. 9 has zero remanence
within the CT core. The distortion of the secondary current
starts at the second cycle after the fault occurs, one cycle
earlier than the case shown in Fig. 8. In Fig. 10, the
remanent flux is 20% $,, the secondary current is also
distorted at the beginning of the second cycle, but the
501
 cycle of secondary current waveform, including transform-
..... Compensated sec. ing the waveform using db4, detecting the distorted section
from healthy section, and then compensating the waveform
9 5 with the regression technique, was 10ms. In a more
$! practical hardware platform, the processing speed is
: o dependant on the speed of the embedded microprocessor
and sampling rate, and the entire signal processing speed is
likely to be reduced even further with a high speed
-5
processor and appropriate sampling rate.
The overall distribution of the errors of the simulated
100 ......... before compensated
. 20 cases is shown via histograms in Fig. 13. It can he
clearly seen that before the compensation the maximum
error reaches as high as 80%. and the average error is
z more than 50%, while after the compensation,
g 50 the maximum error is 3.5%, and the average is within
b
2%. Among all the sampling points after compensation,
there are only 10% of the points whose error falls in the
0 range of 2-3.5%. They mostly appear in cases with extreme
20 40 60 severe conditions.
lime. ms

Fig. 11 As Fig. 8 hut for 60%;reiiiaiienf p u . ~ 2000

I

-5

.........

0 50 0 2 4
before aner
D
i :,
: .
: i
error I" %
Fig. 13 Seconrirrry current error before ond ufrer mmpensution
20 40 60
lime. ms

Fig. 12 A s for Fig. 8 bur 80%;renranentpur

5 Conclusions

waveshape is deformed more severely. With the case shown A new technique to accurately compensate for CT
in Fig. 1 I , when the remanence increases to 60% $.v, the secondary current distortion due to core saturation,
distortion sets in at a much early stage, i.e. the first cycle particularly in the presence of remanent flux when the
after the occurrence of the fault. distortion is most severe has been demonstrated. The
Fig. 12 illustrates the distortion associated with 80% technique first utilises the wavelet transform to accurately
remanent flux; this is the maximum flux that is possible distinguish the distorted section from the healthy section of
from a practical CT. Under this situation the transit flux has a current waveform; secondly, the technique uses regression
a very short effective excursion before saturation ensues. It to recover the distorted section caused by saturation
is apparent from the Figure that the distortion of the essentially to restore it to the requisite scaled down replica
secondary current is the severest, and sets in only a few of the primary system current.
milliseconds after the occurrence of a fault. The effectiveness of the proposed technique has been
The cases presented cover a wide range of fault tested on 20 fault cases on a sample system. These faults
conditions, including the extreme one which is relatively represent a wide spectrum of conditions: the inception angle
uncommon. Despite the wide variety, all distorted sections range from 0 to 180°, and the remanent flux in the CT core
of the secondary currents are satisfactorily detected and from -75 to 80% of the flux at the saturation point. Despite
compensated for by the proposed technique. the wide variations. the technique is able to recover the
The simulations throughout this study were carried out distorted secondary current from an average error of 50%
on a 800MHz Celeron PC with the MATLAB 5.2 and a maximum error of 80% before compensation, to 2
platform. A typical processing time to process a complete and 3.5%, respectively, after the compensation. thereby
502 I€€ Proc.-Gmcr. Trrmrm. Di.wr6. Vo!. 149, No. 4. July 2G92
 ~

clearly demonstrating the effectiveness and robustness of 7 LUKOWICZ, M., m d ROSOLOWSKL E.: 'Artificial neural
the proposed technique. network based dynamic compensation of current transformer error'.
Proceedings of the 8th intcmational symposium on sharisircuit
currents in pou'er systems, B N S ~ S Belgium,
. October 1998. pp. 19-
Acknowledgment 24
8 PIHLER, 1.. GRCAR, B., and DOLINAR. D:. 'Improved operation
oipower transformer protection using artificial neural network'. IEEE
The authors would like to thank Nuffield Foundation for Tronr Power Ddiu, 1999, 12, (3), pp. 1128-1135
financial support of this work. 9 KANG, Y.C., PARK, J.K., K A N F . S.H.. JONES. A.T.. and
AGGARWAL, R.K.: 'An algorithm for compensation secondary
currents of currcnt transformers'. IEEE T r m . Poxer Dcliv.. 1997. 12.
7 References (I), pp. 116124
10 ATP Rule Book. Canadian/Americdn EMTP user Faup. 1992
I WRIGHT. A,. and CHRISTOPOULOS, C.: 'Electrical power system
protection' (Chapman & Hall. 1993)
2 LEWIS BLACKBURN. J.: 'Protective relaying' (Marcel Dekker.
- I I KEZUNOVIC. M., FROMEN. C.W., and PHILLIPS, F.: 'Expen-
mental evaluation of EMTP-based wrrrnr transformer models for
orotectivc rdav trmsient studv'. IEEE Trans. P o w r Ddir.. 1Y94. 9.
19x7)
3 12
4 13
5
compensation of current transformers', IEEE Trum Power Appm 14
SJSI.. 1978.97, (4), pp. 1264-1271 tool for the resonant grounded power dirtnbulion systems relaying'.
6 XU. 0..REFSUM, A,. and WATSON, R.: 'Application ofextemsl IEEE Trani. POIWDelio., 1496. II. (3). pp. 1301-1308
campensarion to current Iransformers'. IEE Proc..Sci. Meas. Techno/.. li HILDEBRAND, F.B.: 'Introduction to numencal amxlysis'
1996. 43, (2), pp. 147-150 (McGraw-Hill. 1974, 2nd Edn.)

503