Electrical Power and Energy Systems 43 (2012) 1376–1382

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

A novel method for adaptive distance protection of transmission line connected

to wind farms
Hadi Sadeghi
Department of Electrical Engineering, Amirkabir University of Technology, Tehran, Iran

a r t i c l e i n f o a b s t r a c t

Article history: Wind speed variations results in wind farm voltage, frequency and power output fluctuations. Therefore,
Received 8 July 2010 protection of lines connecting such a farm to the grid is very important and an adaptive system for dis-
Received in revised form 17 June 2012 tance protection of such a line is necessary. In this paper, an adaptive unit which adjusts the relay trip
Accepted 26 June 2012
characteristic using local information has been designed for distance relay using artificial neural net-
Available online 31 July 2012
works. In this case, in order to prevent wrong operation of relay, changing in wind farm conditions,
the set points of different zones of distance relay has to be changed simultaneously. The results obtained
Keywords:
from proposed method are verified by computer simulation.
Adaptive protection
Digital protection
Ó 2012 Elsevier Ltd. All rights reserved.
Distance relaying
Wind farm

1. Introduction study. With an adaptive feature in a distance relay, the boundary

is set online in accordance with the prevailing condition. In [6],
Wind farms are increasingly integrated to the grids at different the trip boundary is set adaptively that assumes that through SCA-
levels of voltage across the world. The share of such farms in a DA or PMU voltages and line flows of all parts of the system are
power system is also rising day by day. The problem that arises in available. The module of adaptive relay using the swiveling quadri-
integrating such farms is primarily due to variable wind speed. lateral characteristics that the angle of swivel is computed using
The speed variation results in fluctuating output power. The output residual current where the fault area is assumed to be fixed, is pro-
power of a generating unit has a nonlinear relationship with the posed in [7]. The proposed scheme in [8] is based on neural network
wind speed, and when the speed is beyond the limits, the farm can- using real and reactive power at the relay location as the input vec-
not contribute to the grid [1]. In addition, due to under/over voltage tor. Such an approach provides an approximate solution and the
conditions, a group of turbines may trip while others may remain in neural network is not valid for another system.
operation. Obviously, the transmission system that connects such In adaptive scheme which is proposed in this paper, the ratio of
farms will be exposed to such a continuously changing environ- local current and voltage and instant information of wind farm has
ment. In most papers, topics related to protection of wind farms been used for adaptive protection. The proposed approach is sim-
have discussed about over-current relay setting in distribution sys- ple to implement and provides accurate settings for such a system.
tems and adaptive schemes are proposed for distribution systems Results are provided for a line-to-ground fault and the concept can
connected with wind generators [2,3]. The protection of transmis- be extended to other types of faults. The trip characteristic of a re-
sion line connected to wind farms is discussed in [4]. Distance re- lay is decided from detailed off-line study of the system. In an
lays are commonly used for line protection either as primary or adaptive form, the trip boundary should be changed as the system
backup. Their digital version has advantages of better monitoring, condition change. In [6], the mathematical formulation is outlined
communication, and adaptation to system condition. Adaptive for generating trip boundary in the case of line-to-ground fault.
forms of distance relays are proposed to overcome associated prob- The trip boundary considered here is of quadrilateral characteristic
lems in real time, which ultimately increase the overall reliability on an impedance plane obtained by varying fault location and fault
index of the protection scheme [5–13]. In distance protection, to resistance within their limits.
coordinate high fault resistance with ground faults, quadrilateral
characteristic is preferred. In a fixed setting approach, the boundary
of the relay characteristic is predefined based on overall system 2. Ideal trip characteristic for distance relay

In a wind farm, a number of units are connected in parallel to

E-mail address: [email protected] harness bulk amounts of electric power for a grid. When such a

0142-0615/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2012.06.072
 H. Sadeghi / Electrical Power and Energy Systems 43 (2012) 1376–1382 1377

potentially high level of generation is available at a remote place,

transmission systems are hooked up for efficient power manage-
ment. As the generation at a wind farm fluctuates, the connecting
line will see varying degrees of transmitted power through a day.
The fluctuations of voltage and frequency in the system depend
on the proportionate strength of the farm. To see the effects of
wind farm operating conditions on trip characteristic, a 400 kV,
60 Hz power system is studied later [6].
A schematic of a typical wind farm connection is shown in
Fig. 1. In this figure, the wind farm supplies power to the grid Fig. 2. Line diagram for phase A-to-ground fault [14].
through the transmission network that is protected by a distance
relay. The line diagram of the power system for phase-A-to-ground
fault in the line is shown in Fig. 2. The considered distance relay is Table 1
positioned at W and the normal power flow direction in the system Values of different parameters of studied system.

is from W to P (wind farm to grid) as marked in the figure. In this ZSW (X) ZSP (X) ZL (X)
analysis, the wind farm and the grid are considered with their Zero sequence 30\85° 1.5\85° 87.35\83°
equivalent simplified models and shunt capacitance is not consid- Positive sequence 20\85° 1\85° 28.75\86°
ered for the line.
The system parameters are presented in Table 1.
The pre-fault voltage relation of the equivalent sources is de-
30
fined as
R-X Plane
jd 25
EAP ¼ qe EAW ð1Þ
The pre-fault current can be written as 20
II
IWP ¼ ðEAW  EAP Þ=Z 1
X (ohm)
ð2Þ 15

I
where Z1 is the positive sequence impedance of the system and can 10
be obtained as
III
5
Z 1 ¼ Z 1SW þ Z 1L þ Z 1SP ð3Þ
From voltage and current relations in (1) and (2), the following 0
can be written IV
-5
0 10 20 30 40 50 60 70 80
qejd ¼ 1  ðIWP =EAW ÞZ 1 ð4Þ R (ohm)
If system conditions (q, d, ZSP, ZSW) are fixed and RF and fault Fig. 3. Ideal trip characteristic for Z1SW = 20\85°, Z0SW = 30\85°, q = 0.95, d = 20.
location varied, four boundary lines, defined below, are obtained
by computer simulation (Fig. 3).

Line I: solid faults at different points of line. 3. Proposed adaptive protection scheme
Line II: faults at relay-reach end (90% of line length) with differ-
ent fault resistance up to 50 X. To prevent mal-operation of distance relay, it is necessary that
Line III: faults at different points with a 50 X fault resistance. relay trip decision can change by variations of wind farm condi-
Line IV: faults at the relaying point with different fault resis- tions. In fact, for a point in firs zone of relay (90% of line length),
tance up to 50 X. impedance value seen by relay must be included inside of trip
characteristic. Thus, the four lines which determines relay trip
These four lines and included area define ideal trip region of boundaries must be varied with wind farm conditions (the amount
relay under the prevailing system conditions. of IWP/EAW and equivalent impedance of the wind farm). Therefore,
In [14], it is shown that variation of two parameters as wind how to make a distance relay operate sensitively as these condi-
farm local information affects on seen impedance by the distance tions vary is the main problem that must be solved. To solve this
relay and ideal trip characteristic. These parameters are equivalent problem, two different approaches can be proposed:
impedance of the wind farm (Z1SW) and the ratio of current and
voltage in relay point (IWP/EAW). (1) Firstly, setting patterns under all possible system conditions
can be calculated off-line and stored in a data table for using
during fault. Alternatively, only one set of setting parameters
are calculated on-line and they are renewed from time to
time as the system conditions change. When a fault occurs
the relay will operate with the latest setting parameters, an
optimal trip boundary [6].

This method is faster in on-line processing due to less computa-

tion. However, a large amount of computer memory would be
needed to cover all the possible conditions. Also, the nonlinear
Fig. 1. Interconnection of wind farm to the grid [14]. boundaries will make it more complex.
1378 H. Sadeghi / Electrical Power and Energy Systems 43 (2012) 1376–1382

Relay setting is the other important problem. Mho relay is the

simplest kind of relay can be used. In this case the ideal trip char-
acteristic which is a quadrilateral shape with nonlinear boundaries,
are surrounded in circle of mho characteristic. However, since the
included area of ideal characteristic is very less than its environ-
ment circle surface, particularly for large amount of Z1SW, and dif-
ferent zones of relay may intervene together, using of mho relay
does not seem reasonable. Also, this relay may trouble in over-
reach or under-reach problem in presence of fault resistance and
thus, quadrilateral relay should be used instead of mho relay.
Ideal trip characteristic can be fitted in a real quadrilateral dis-
tance relay characteristic in an optimized manner as shown in
Fig. 4.
But this work has the problem of intervention of shaded area
with the second zone of relay. If the shaded area can be deleted,
Fig. 5. Adaptive setting characteristic and topological trip-decision logic [6].
the worry of invention of zone-1 and zone-2 will be obviated. For
this purpose, described approach in the following will be preceded
to obtain zone-1 trip decision.

(2) Adaptive scheme: the computation of the adaptive setting

scheme consists of two parts:
(I) The computation of the setting boundaries.
(II) The calculations performed during a fault.

This scheme which has been carefully designed, minimize cal-

culation of both parts while maintaining high reach accuracy. In
this scheme, a number of straight line segments are used to
approximate the curved boundaries in Fig. 3. The number of the Fig. 6. Calculation architecture in adaptive setting [6].
straight lines is determined by the precision required and process-
ing capacity of the processor. Here, three straight lines 2, 4, 6 are
used for line II, two 8, 10 for line III, two 12, 13 for line IV and line
1 for line I. This approximate model is shown in Fig. 5.
Each line divides the R–X plane into two region and we suppose
that a ‘‘line output’’ is positive if the measured impedance falls on
one side of the line and negative on the other side. The positive or
zero value of line output is assumed comparable with logical ‘‘1’’
and negative value with ‘‘0’’. If the ideal trip region remained a con-
vex shape these linear outputs and some ‘‘AND’’ operations could
well determine a trip decision during fault. However, since differ-
ent system conditions will result in different curved boundaries,
the decision region may be non-convex. Hence mere logical
‘‘AND’’ operations of the linear outputs cannot be relied upon for
the trip decisions. For this reason, additional lines are used to di-

Fig. 7. Trip decision logic circuit [6].

vide the decision region into several sub-regions. These lines are
shown with numbers 5, 7, 9, 11 and 13 in Fig. 5. The linear outputs
for straight lines 1–13 in Fig. 5 correspond to signals L1–L13 in
Fig. 6.
Now, a combination of logical ‘‘AND’’ and ‘‘OR’’ operations of
these signals can be used to obtain reliable trip decision. Each lin-
ear output is connected to the measured reactance and resistance
through two weights. A bias weight is also connected to the linear
output. For example, for L1 we have W11 and W12 and also bias
weight of W10. The output value is defined by

L1 ¼ W 11 X A þ W 12 RA þ W 10 ð5Þ
Positive and negative values of the output determine the two
sides of the line. A critical thresholding condition occurs when
the linear output equals zero:

W 11 X A þ W 12 RA þ W 10 ¼ 0 ð6Þ
Fig. 4. Optimized trip characteristic for distance relay and ideal trip characteristic. Therefore:
 H. Sadeghi / Electrical Power and Energy Systems 43 (2012) 1376–1382 1379

Fig. 8. Position of selected points for polygonal trip characteristics corners.

Fig. 10. Performance function of 3rd network.

Fig. 9. 3rd Neural network diagram. line, and selecting W11 = ±1 (to minimize on-line calculation during
faults), the three weights are:
W 10 ¼ 0
X A ¼ ðW 12 =W 11 ÞRA þ ðW 10 =W 11 Þ ð7Þ W 11 ¼ 1 ð8Þ
It is obvious that the above equation expresses a straight line W 12 ¼ X A ðaÞ=RA ðaÞ
having slope and intercept given by (W12/W11) and (W10/W11)
A similar method is used to obtain weights for L3, L5, L7, L9, L11
respectively and, therefore, the three weights. W10, W11 and W12,
and L13. However, it is only slightly more complex to determine
determine the slope, intercept, and side of separating line.
the weights for L2, L4, L6, L8, L10 and L12. Taking L2 as an example
For example, in the case of line ‘‘1’’ in Fig. 5 the intercept is zero
we have:
and the slope can be determined by resistance and reactance at ‘‘a’’
only. Supposing the linear output is positive on the left side of the L2 ¼ W 21 X A þ W 22 RA þ W 20 ¼ 0 ð9Þ

Table 2
The values of resistance and reactance of 3rd point for different condition of system.

Z1SW = 20\85° Z1SW = 40\85° Z1SW = 60\85° Z1SW = 80\85° Z1SW = 20\85°
q = 0.9, d = 3° (94.59, 43.83) (124.57, 48.74) (154.07, 53.65) (183.33, 58.56) (212.45, 63.47)
q = 0.9, d = 6° (87.30, 36.77) (115.38, 39.83) (143.04, 42.92) (170.47, 46.03) (197.78, 49.14)
q = 0.9, d = 9° (80.75, 31.35) (107.10, 32.87) (133.07, 34.46) (158.83, 36.07) (184.48, 37.69)
q = 0.9, d = 12° (74.89, 27.12) (99.66, 27.36) (124.08, 27.68) (148.32, 28.04) (172.45, 28.42)
q = 0.9, d = 15° (69.66, 23.78) (92.98, 22.93) (115.98, 22.19) (138.83, 21.49) (161.57, 20.83)
q = 0.9, d = 18° (64.98, 21.10) (86.97, 19.33) (108.67, 17.68) (130.24, 16.09) (151.71, 14.54)
q = 0.9, d = 21° (60.78, 18.93) (81.54, 16.37) (102.06, 13.95) (122.44, 11.59) (142.75, 9.27)
q = 0.9, d = 24° (57.00, 17.15) (76.63, 13.92) (96.05, 10.82) (115.35, 7.80) (134.58, 4.82)
q = 0.9, d = 27° (53.58, 15.68) (72.16, 11.85) (90.56, 8.18) (108.86, 4.58) (127.10, 1.02)
q = 0.9, d = 30° (50.46, 14.44) (68.08, 10.11) (85.54, 5.93) (102.91, 1.82) (120.22, 2.23)
q = 1.0, d = 3° (108.29, 23.94) (141.87, 24.13) (174.92, 24.40) (207.71, 24.71) (240.34, 25.04)
q = 1.0, d = 6° (97.44, 19.72) (128.49, 18.55) (159.05, 17.50) (189.37, 16.50) (219.54, 15.53)
q = 1.0, d = 9° (88.42, 16.66) (117.22, 14.38) (145.60, 12.25) (173.74, 10.19) (201.75, 8.16)
q = 1.0, d = 12° (80.80, 14.38) (107.62, 11.19) (134.05, 8.17) (160.28, 5.22) (186.38, 2.32)
q = 1.0, d = 15° (74.30, 12.65) (99.34, 8.70) (124.04, 4.93) (148.55, 1.25) (172.96, 2.37)
q = 1.0, d = 18° (68.67, 11.29) (92.12, 6.71) (115.26, 2.32) (138.25, 1.97) (161.14, 6.21)
q = 1.0, d = 21° (63.76, 10.23) (85.77, 5.11) (107.51, 10.18) (129.12, 4.63) (150.65, 9.40)
q = 1.0, d = 24° (59.43, 9.38) (80.13, 3.80) (100.61, 1.57) (120.97, 6.85) (141.25, 12.07)
q = 1.0, d = 27° (55.58, 8.69) (75.10, 2.72) (94.42, 3.05) (113.64, 8.73) (132.79, 14.34)
q = 1.0, d = 30° (52.13, 8.13) (70.56, 1.81) (88.83, 4.30) (107.01, 10.32) (125.12, 16.28)
q = 1.1, d = 3° (112.74, 0.16) (148.44, 5.00) (183.51, 10.01) (218.26, 14.95) (252.83, 19.86)
q = 1.1, d = 6° (100.54, 0.48) (133.28, 5.37) (165.46, 11.02) (197.37, 16.56) (229.10, 22.04)
q = 1.1, d = 9° (90.64, 0.76) (120.82, 5.63) (150.51, 11.78) (179.96, 17.81) (209.25, 23.77)
q = 1.1, d = 12° (82.44, 1.01) (110.38, 5.63) (137.90, 12.37) (165.20, 18.80) (192.38, 25.16)
q = 1.1, d = 15° (75.52, 1.24) (101.49, 5.93) (127.11, 12.84) (152.53, 19.61) (177.84, 26.31)
q = 1.1, d = 18° (69.61, 1.44) (93.83, 6.02) (117.76, 13.22) (141.51, 20.29) (165.17, 27.27)
q = 1.1, d = 21° (64.48, 1.62) (87.15, 6.09) (109.56, 13.54) (131.83, 20.85) (154.02, 28.08)
q = 1.1, d = 24° (59.99, 1.79) (81.26, 6.13) (102.32, 13.80) (123.25, 21.33) (144.10, 28.78)
q = 1.1, d = 27° (56.02, 1.94) (76.03, 6.16) (95.86, 14.02) (115.57, 21.74) (135.23, 29.39)
q = 1.1, d = 30° (52.48, 2.07) (71.34, 6.19) (90.05, 14.21) (108.67, 22.10) (127.23, 29.92)
1380 H. Sadeghi / Electrical Power and Energy Systems 43 (2012) 1376–1382

30 Table 4
Value of seen impedance in relay point for different fault resistance occurred in 60% of
line length.
25
RF = 10 X RF = 20 X RF = 30 X RF = 40 X RF = 50 X RF = 60 X RF = 70 X

20 18.99 + j 32.48 + j 43.31 + j 52.17 + j 59.56 + j 65.80 + j 71.15 + j

18.83 11.9 10.73 10.02 9.59 9.36 9.24
X (ohms)

15

10 30

New setting for:

2 Z1sw=1.74+19.92i
5 25
I/E = 0.0067-0.0016i

0 20
1

-5 15

X (ohms)
0 20 40 60 80 100 120 40 ohms

R (ohms) 10
10 ohms
20
Fig. 11. (1) Ideal trip characteristic and (2) approximate characteristic obtained ohms
30
70
from neural networks output. 5 ohms 50 60
ohms
ohms ohms

0
Table 3
Value of seen impedance in relay point for different position of fault on-line length
and RF = 10 X. -5
0 10 20 30 40 50 60 70 80 90 100
L = 30% L = 50% L = 70% L = 90% L = 98% R (ohms)
11.71 + j 15.83 + j 23.66 + j 56.33 + j 82.45 + j
7.095 11.78 15.51 15.32 14.4 Fig. 13. Seen impedances from relay point for faults occurred in 60% of line length
and different fault resistances (Z1SW = 20\85°, IWP/EAW = 0.0067j 0.0016).

30 Table 5
The weight values comparable with trip characteristic presented in Fig. 3.

25 i Wi0 Wi1 Wi2

90% 1 0 1 9.909
Line 98% 2 26.049 1 0.186
20 Line 3 0 1 0.219
4 18.083 1 0.064
15 5 0 1 0.144
X (ohms)

6 13.667 1 0.013
7 0 1 0.125
10 70% 8 5.501 1 0.069
Line 9 0 1 0.162
50%
Line 10 7.178 1 0.283
5 11 0 1 0.113
30% 12 22.187 1 0.638
New setting for:
0 Line 13 0 1 0.043
Z1sw =1.74+19.92i
I/E = 0.0067 - 0.0016i

-5
0 10 20 30 40 50 60 70 80 90 100 During fault, after fault detection and fault classification, a dig-
R (ohms) ital distance relay calculates the apparent reactance and resistance.
From these apparent values the linear outputs L1–L13 can be
Fig. 12. Seen impedances from relay point for faults occurred in different points of obtained. In proposed adaptive scheme, the logical values compa-
transmission line and Z1SW = 20\85°, IWP/EAW = 0.0067j 0.0016.
rable with positive or negative value of these outputs using
‘‘AND’’ and ‘‘OR’’ operations between them are used to reach a final
trip decision as shown in Fig. 7.
Therefore:
It is obvious that the computational task for the processor is
W 21 X A ðaÞ þ W 22 RA ðaÞ þ W 20 ¼ 0 small. In this study case, the post-fault online calculations are no
ð10Þ more than 13 multiplication, 19 additions and 12 logic operations.
W 21 X A ðbÞ þ W 22 RA ðbÞ þ W 20 ¼ 0
This involves relatively little real time digital processing and can be
and supposing that the linear output is required to be positive on well done by 16-bit microprocessors [6].
the trip region side and selecting W21 = 1, the weights are given Evidently, the resistance and reactance values of points ‘‘a’’
by: through ‘‘g’’, all depends on Z1SW and IWP/EAW. Therefore, to obtain
them, seven FFB1 neural networks are used here. The inputs of each
W 21 ¼ 1
W 22 ¼ ðX A ðbÞ  X A ðaÞÞ=ðRA ðbÞ  RA ðaÞÞ ð11Þ
W 20 ¼ X A ðbÞ  RA ðbÞðX A ðbÞ  X A ðaÞÞ=ðRA ðbÞ  RA ðaÞÞ 1
Feed Forward Back propagation.
 H. Sadeghi / Electrical Power and Energy Systems 43 (2012) 1376–1382 1381

Fig. 14. Block diagram of distance relay with the adaptive setting scheme [6].

of them are Z1SW and IWP/EAW and the outputs are resistance and is positioned on 60% of line length. Table 4 represents the mea-
reactance of one of the points ‘‘a’’ through ‘‘g’’. The number of inputs sured impedance values.
and outputs of each network are 2. If other trip characteristic were It can be seen easily from Fig. 13 that increasing fault resistance,
used for distance relay, the required number of outputs might be less the locus of seen impedance do not track complete horizontal
or more than this case. In fact, the number of network outputs will direction and the measured impedance drop inside of trip charac-
be depended on the number of variables required to determine relay teristic for the faults up to 50 X. This adaptive method, improves
trip region. The reason that we used seven networks with two out- the protection system treatment against fault resistance variation
puts instead one network with 14 outputs, is to increase their reli- because a polygonal characteristic treatment is better than mho
ability and learning speed and to decrease their total error. characteristic against fault resistance variations.
To train the networks, for each point, 150 cases are acquired by Now, it is the time that the presented weights in previous sec-
changing in parameters which expressed above. The resistance and tion for lines L1–L13 have to be obtained from neural networks
reactance value of each point is acquired by changing in RF and output points. The results can be observed in Table 5.
fault location as shown in Fig. 8. After the weights and positive or negative value of linear out-
To define and train each network, we used MATLAB neural net- puts for online seen impedance was calculated, the adaptive unit
work toolbox. The kind of networks is perceptron and they were obtains a trip decision.
trained by back-propagation method. The neural transfer function Extra high speed digital distance relays have been implemented
of input layer is tangent sigmoid and of output layer is linear. Usu- with different algorithms and some can operate within one cycle or
ally, the number of input layer neurons is determined according to even a few milliseconds of fault occurrence. A typical digital dis-
performance function of network and experimentally. For example, tance relay consists of a fault detection unit, a fault classification
here, this number is considered 150 for 3rd network. Training func- unit, a fault measurement unit and a trip region comparison unit.
tion is ‘‘scaled conjugate gradient back propagation’’.2 A disturbance or a fault is first detected and then the fault type
The diagram of 3rd neural network is shown in Fig. 9. This net- is determined. Apparent reactance and resistance are calculated
work was trained using values presented in Table 2 and obtained in the measurement unit. Generally, successive estimates of the
from relation 4. Its performance function curve is shown in Fig. 10. impedance are compared with the boundary conditions at every
sampling interval. To improve the performance of currently avail-
4. Testing of proposed scheme and results able digital distance relays a practical scheme shown in Fig. 14 was
designed.
Since Z1SW and IWP/EAW have various values, trained networks In the block diagram, an adaptive setting (AS) unit works in col-
must be tested to ensure from their proper performance. For exam- laboration with a default setting (DS) unit and the default setting
ple, suppose that Z1SW and IWP/EAW is equal to 20\85° and unit with a preset conventional quadrilateral characteristic re-
0.0067  j 0.0016 respectively. Entering these values to neural net- mains valid during a fault until its recovery. The adaptive setting
works program, the points ‘‘a’’ through ‘‘g’’ will be acquired. (AS) will be unblocked for a certain period during a fault only.
The ideal trip region and the characteristic formed by outputs of One reason for this is that the accuracy of the measured impedance
networks, has been shown in Fig. 11. It is evident from this figure is adversely affected by transient components in the first cycle of a
that approximately all of the output points of networks (specified fault and using the adaptive setting with an almost whole line
with stars) are located on ideal trip characteristics. This subject ap- boundary may cause mal-operation. Another reason is that false
proves that neural networks have very good accuracy. tripping may also be caused by the removal of the remote-end
To test the performance of this adaptive unit, computer simula- infeed.
tion for a fault located on 90% of line length has been done. Basi-
cally, the measured impedance in relay point must be positioned 5. Conclusion
inside of the trip characteristic. For longer distances the impedance
must be positioned outside of that region. Table 3 shows the fault Power output of a wind farm fluctuates throughout a day and
location and seen impedance value in relay point. The fault resis- the transmission system connecting such a farm to the grid, will
tance has been considered equal to 10 X. be exposed to these power variations. The simulation results for
In Fig. 12, the adaptive trip characteristic and the location of a two-terminal line which connects the wind farm to the grid,
impedances of Table 3 have been shown, and the relay perfor- show that the wind farm conditions, including the number of gen-
mance can be discovered from this figure. As it is obvious from erating units and the ratio of local current and voltage affect the
Fig. 12, for a fault located on distances shorter than 90% of line trip boundary of the distance significantly. To take into account
length, the proposed adaptive unit will detect it properly and it will the changing wind farm condition, an adaptive setting method
not render trip command for longer distances. for different zones of distance is proposed. In method discussed
Now, for the same values of Z1SW and IWP/EAW, we change the in this paper, a separate adaptive unit that works in collaboration
fault resistance and measure impedance in relay point. The fault with a default setting unit and uses local information of wind farm
has been used and the output points of some neural networks, con-
struct the characteristic of this unit. Using of neural network in
2
TRAINSCG. adaptive setting has a good accuracy, reliability and also high
1382 H. Sadeghi / Electrical Power and Energy Systems 43 (2012) 1376–1382

speed because the number of online measurements is limited in it [6] Xia YQ, Li KK, David AK. Adaptive relay setting for standalone digital distance
protection. IEEE Trans Power Del 1994;9:480–91.
and the need of large amount computer memory has been solved.
[7] Moore PJ, Aggarwal RK, Jiang H, Johns AT. New approach to distance protection
The proposed adaptive scheme in this paper does make a digital for resistive double-phase to earth faults using adaptive techniques. IEEE Proc
distance relay operates with improved sensitivity, selectivity and Gen Trans Distrib 1994;141(4):369–76.
speed. [8] Li KK, Lai LL, David AK. Stand alone intelligent digital distance relay. IEEE Trans
Power Del 2000;15:137–42.
[9] Sidhu TS, Baltazar DS, Palomino RM, Sachdev MS. A new approach for
References calculating zone-2 setting of distance relays and its use in an adaptive
protection system. IEEE Trans Power Del 2004;19:70–7.
[1] Erikson PB et al. System operation with high wind generation. IEEE Power [10] Orduna E, Garces F, Handschin E. Algorithmic-knowledge-based adaptive
Energy Mag 2005;3(6):65–74. coordination in transmission protection. IEEE Trans Power Del 2003;18:61–5.
[2] Jang SI, Choi JH, Kim JW, Choi DM. An adaptive relaying for the protection of a [11] Izykowski J, Rosolowski E, Saha M. Adaptive digital distance algorithm for
wind farm interconnected with distribution network. In: Proc IEEE PES trans parallel transmission lines. In: Proc IEEE PowerTech conf, vol. 2; 2003.
dis conf expo, vol. 1; 2003. p. 296–302. [12] Jonsson M, Daalder JE. An adaptive scheme to prevent undesirable distance
[3] Brahma SM, Girgis AA. Development of adaptive protection schemes for protection operation during voltage instability. IEEE Trans Power Del
distribution systems with high penetration of distribution generation. IEEE 2003;18:1174–80.
Trans Power Del 2004;19:56–63. [13] Dutra R, Fabiano L, Oliveira W, Saha MM, Lindstorm S. Adaptive distance
[4] Hornak D, Chau NHJ. Green power-wind generated protection and control protection for series compensated transmission line. In: IEEE/PES trans distrib
consideration. In: Proc IEEE 57th annu conf protective relay eng; 2004. p. 110– conf expo; 2004. p. 581–6.
31. [14] Pradhan AK, Joos G. Adaptive distance relay setting for lines connecting wind
[5] IEEE Guide for protective relay applications to transmission lines. standard farms. IEEE Trans Energy Conv 2007;22:206–13.
C37.113-1999, February 29, 2000.