Electrical Power and Energy Systems 49 (2013) 243–252

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems

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

An ANFIS-based fault classification approach in power distribution system

J. Zhang a, Z.Y. He b,⇑, S. Lin b, Y.B. Zhang a, Q.Q. Qian b
a
State Grid Energy Research Institute, Beijing, China
b
Department of Electrical Engineering, Southwest Jiaotong University, Chengdu, SC, China

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

Article history: Fault classification is very important for power system operation because it is the premise of fault
Received 5 August 2010 analysis process. In this paper, an ANFIS (Adaptive Neural Fuzzy Inference System) based fault classification
Received in revised form 11 December 2012 scheme in neutral non-effectively grounded distribution system is proposed. The transient currents are
Accepted 20 December 2012
obtained by wavelet transform after faults occur. According to the statistic characteristic of transient
Available online 26 February 2013
currents in different fault types, the fault identifiers are defined. The fault identifiers can characterize
the traits of fault type and show different disciplinarian in different fault types. They are inputted into
Keywords:
three ANFISs to obtain the fault type. The proposed approach only needs the voltages and currents
Fault classification
Neutral non-effectively grounded
measured at substation, and can identify ten types of short-circuit fault accurately. The simulation model
distribution is established in PSCAD/EMTDC environment, and the performance of the proposed approach is studied.
Adaptive Neural Fuzzy Inference System The results show that it has high accuracy. Besides, the adaptability of proposed approach to the neutral
(ANFIS) compensated grounding system, different network configurations and so on are verified through
Wavelet transform simulation. Through simulation, the proposed approach exhibits good performance.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction power systems, the acquisitions of transient electric signals are

Once fault occurs in power system, fast and accurate fault
classification is very important for post-fault analysis and power
supply restoration. On the one hand, fault type information is the
premise of fault location. Correct fault classification helps fault
diagnosis system select appropriate fault location scheme. On the
other hand, as the automation of distribution network advances,
a mass of data is uploaded after fault. So, it is impossible for oper-
ators to classify fault correctly by their experience.
A lot of work has been focused on the fault classification prob-
lem in transmission network [1–6]. It is probably due to the impor-
tance of fault classification for relay operations. As the amount of
power carried by distribution networks increases during the past
decades, the need of effective fault classification scheme becomes
more and more necessary for operators. The research of fault clas-
sification in distribution network can be summarized to two cate-
gories approximately: (1) the schemes using stationary electric
components [7–9], and (2) the schemes using transient electric
components [10,11]. The method proposed in [7] is widely used
in industry. It utilizes the angular relationships among stationary
components to identify fault type in isolated distribution. The
angular relationships are defined by fuzzy membership functions
in [8], afterward, the fault type is obtained by logic reasoning
mechanism. As fast sampling devices are installed in modern
⇑ Corresponding author. Tel.: +86 2887602445; fax: +86 2887600861.
E-mail address: hezy@swjtu.cn (Z.Y. He).
possible. It highly supports the transient-based post-fault analysis
techniques, such as fault location [10], fault line identification [12]
and so on. In the paper [10], the transient currents are obtained by
stationary wavelet transform. Afterward, by comparison with
energies of transient currents, the fault type is identified.
In this paper, an ANFIS-based fault classification approach in
distribution network is proposed. It uses transient fault signals to
actualize ten types of short-circuit fault classification (AG, BG,
CG, ABG, ACG, BCG, ABC/ABCG, AB, AC and BC). The whole classifi-
cation process contains 3 steps: the transient fault signals are ex-
tracted by WT (Wavelet Transform) at first. At second, the
extracted signals are calculated for the statistic quantities, which
are called FIs (Fault Identifiers) here. At third, the FIs are inputted
into three ANFISs (Adaptive Network-based Fuzzy Inference Systems)
to obtain the final result. The simulation model is established in
PSCAD/EMTDC environment. The classification accuracy of pro-
posed approach is verified under different fault circumstances.
Moreover, the adaptability of proposed approach to different dis-
tribution operations is thoroughly studied. The results show that
its adaptability is good.
The remainders of this paper are constructed as follows. The
whole structure of proposed approach is introduced in section 2
for clearness. The regulars of FIs in different fault types are studied
in section 3. The ANFIS-based fusion process is introduced in sec-
tion 4. The classification accuracy of proposed approach is exhib-
ited in section 5. The adaptability of proposed approach is
studied in section 6. At last, the conclusion is given in section 7.

0142-0615/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2012.12.005
244 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252

2. The whole structure the coefficients of high-pass filter in wavelet decomposition and
reconstruction calculation respectively.
00
Although it is important how to trig fault classification The current ip is implemented 3 levels of wavelet decomposi-
approach, it is not the certain in this paper. The method proposed tion. The detail coefficients of 2 and 3 levels are reconstructed.
0
in the paper [10] can be employed to trig the fault classification The reconstructed signal is denoted by ip . As long as the sampling
0
approach and find the fault inception time. The structure of rate is 10 kHz, it can be known that the frequency band of ip
classification approach is depicted in Fig. 1. ranges from 625 Hz to 2.5 kHz. The frequency band employs tran-
In Fig. 1, the inputs of fault classification are the zero-sequence sient components as much as possible, while avoids the interfer-
00
voltage u0(t) and the fault-components of three-phase currents ia , ence of the 3rd–11th harmonics which are normal in power
00 00
ib and ic . They are calculated by formulas (1) and (2), system.
Fig. 2 shows the original currents and WT-extracted currents in
u0 ðtÞ ¼ ua ðtÞ þ ub ðtÞ þ uc ðtÞ; ð1Þ ACG fault occurring at 0.1 s.
00
ip ðtÞ ¼ ip ðtÞ  ip ðt  TÞ: ð2Þ It can be seen from subfigure (b) in Fig. 2 that the WT-extracted
signals of faulty phases (phase-a and -c) vary more intensively than
where p stands for phase-a, b, and c, ip(t) is the phase-p current in the WT-extracted signal of healthy phase (phase-b). The descrip-
the first basic cycle after fault inception at the secondary winding tion is qualitative, so the quantitative description should be found
of transformer, ua(t), ub(t) and uc(t) are the three bus voltages in 0
out. As long as ip is the signal in one basic cycle, it has 200 sampling
the first cycle after fault inception, and T is the basic cycle (20 ms). 0 0 0
points. There would be 600 inputs if ia , ib and ic are inputted into
00
As can be seen from Fig. 1, u0(t) and ip are preprocessed by FFT ANFISs directly. This is unpractical. Since the statistic quantities
(Fast Fourier Transform) and WT separately. Moreover, the statistic can characterize the shape and energy of signals, six statistic quan-
quantities are calculated for WT-extracted signals. The 2 steps tities are selected from 12 typical statistic quantities [13], and they
above aim at constructing the FIs, which are inputted into ANFISs are calculated by,
to obtain the classification result. The FIs and their fusion process
are very important for fault classification, because FIs must have sp
distinguished features in different fault types, and the fusion pro- sp ¼ p ¼ a; b; c; ð6Þ
smax
cess should fuse FIs effectively to obtain the correct results.
 
 
 0 0 0
Eðia ib Þ  Eðia ÞEðib Þ
0 
3. Fault Identifiers (FIs)  
qa;b ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi; ð7Þ
 0 2 2 0 0 2 2 0 
 Eðia Þ  E ðia Þ Eðib Þ  E ðib Þ
When fault occurs in power system, transient oscillation is
dominant. To utilize the transient electric signals, WT technique  
 
is employed for its flexible time–frequency focus. By comparison  0 0 0
Eðia ic Þ  Eðia ÞEðic Þ
0 
 
with Daubechies series wavelets, the quadratic spline wavelet is qa;c ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi; ð8Þ
 0 2 2 0 0 2 2 0 
selected as the mother wavelet in this work because of its good  Eðia Þ  E ðia Þ Eðic Þ  E ðic Þ
performance in fusion process. The coefficients of wavelet filter gi-  
ven by formulas (3)–(5) are typical, and are adopted in this work.  
 0 0 0
Eðib ic Þ  Eðib ÞEðic Þ
0 
pffiffiffi  
lo ¼ ½ 0:125 0:375 0:375 0:125   2; ð3Þ qb;c ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi: ð9Þ
 0 2 2 0 0 2 2 0 
 Eðib Þ  E ðib Þ Eðic Þ  E ðic Þ
pffiffiffi
hi ¼ ½ 1:0 1:0   2; ð4Þ In formula (6),

hii ¼ ½0:015625 0:109375 0:34375  0:34375 !12

pffiffiffi 1 X n
0 0
sp ¼ ðip ðnÞ  Eðip ÞÞ ; n ¼ 200; ð10Þ
 0:109375  0:015625  2: ð5Þ n  1 i¼1
In (3)–(5), lo is the coefficient of low-pass filter in wavelet
decomposition and reconstruction calculation, and hi and hii are
smax ¼ maxðsp Þ: ð11Þ

Source
Fault classfication

Bus
Construct fault identifiers Fusion
Trans- AG,
former BG,
Ia''(t) CG,
ib''(t) Wavelet Statistic
ABG,
ic''(t) calculation
ACG,
ANFIS BCG,
Bus ABC/
ABCG,
FFT
AB,
u0(t)
AC,
BC.
Feeders

Fig. 1. Structure of fault classification.

 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252 245

100  AG and ABG can be distinguished from each other, because s⁄ of

the faulty phase is bigger than s⁄ of the healthy phases (subfig-
ures (a) and (b)), and q between two faulted phases is bigger
50 than the others (subfigure (b)).
 AB and ABC/ABCG can be distinguished from each other,
Current i (A)

-Phase C
because the s⁄ of the faulty phase is always bigger than that
-Phase B of the healthy phase (subfigures (c) and (d)), and q between
0
two faulty phases is visibly bigger than others in AB fault (sub-
-Phase A
figure (d)) while symmetric fault does not have such regular
(subfigure (c)).
-50

4. Fusion process
-100
0.08 0.1 0.2 In the classification process in section 3, the word ‘‘bigger’’, ‘‘vis-
Time t (s) ibly bigger’’ and ‘‘zero’’ are heuristic and linguistic. So, the fault
(a) classification can be considered as a fuzzy inference problem. In
fuzzy inference, the definitions of membership functions are not
6 an easy work. The changeful fault conditions make their definitions
difficult. The combination of fuzzy logic with an architecture de-
sign of ANN has lead to the creation of ANFIS which has been used
for every plant [14–19]. The ANFISs can adjust the parameters of
membership functions adaptively from history data and take the
Current i (A)

uncertainties (fault resistances, FIAs, etc.) into account well. This

0 is of utmost importance since it is well known that the constant
thresholds for all fault conditions are very difficult to choose.
In this work, three ANFISs are employed for different purposes.
-Phase C They work as Fig. 4 shown.
From Fig. 4, ANFIS1 is used to identify whether fault is AG, BG
-Phase B
CG, LLG (Line-to-Line-to-Ground) or others. If the result from AN-
-Phase A FIS1 is ‘‘LLG’’, ANFIS2 would be used further. If the result is ‘‘oth-
-8 ers’’, ANFIS3 would be used. Because the outputs of ANFISs are
170 200 400 numeral, their corresponding fault type is shown in Fig. 4. The in-
Sampling Points
put vectors of ANFIS1, ANFIS2 and ANFIS3 are C1, C2 and C3 respec-
(b) tively. They are the combinations of FIs and shown as follows,

Fig. 2. Three phase currents, (a) the original, and (b) the WT-extracted. C 1 ¼ ½ sa sb sc U0  ð12Þ

In formulas (7)–(9), E(x) is the mathematical expectation of

C 2 ¼ C 3 ¼ ½ qa;b qa;c qb;c sa sb sc : ð13Þ
variable x.
As can be seen, sa , sb and sc are the normalized standard devia- The inputs and outputs of ANFISs are shown in Fig. 5.
tions of three phases, qa;b , qa;c and qb;c are the correlation coeffi- The general structure of ANFIS with two inputs is shown in
cients between two phases. Fig. 6 [14].
Besides 6 statistic quantities above, the U0 (zero-sequence As can be seen from Fig. 6, ANFIS can be demonstrated in five
amplitude of fundamental component) is calculated by FFT from layers: the first layer actualizes the fuzziness of inputs; the sec-
u0(t). It is used to judge whether the fault is grounded or not. ond layer calculates the firing strength of each rule; the third
The 6 statistic quantities and U0 are defined as FIs. Take the layer normalizes the firing strengths; the fourth layer determines
waves in Fig. 2 for example, the FIs are: sa ¼ 0:788, sb ¼ 0:305, the consequent parameters of the rule; the fifth layer computes
sc ¼ 1, qa;b ¼ 0:595, qa;c ¼ 0:967, qb;c ¼ 0:773. As can be seen, sa the output of the fuzzy system by summing up the outputs of
and sc are bigger than sb , and the qa;c is bigger than qa;b and qb;c . the fourth layer.
To widely study the performance of FIs in different fault types, The feedforward equations of ANFIS in Fig. 6 are shown as
AG, ABG, AB and ABC/ABCG faults are assumed in the isolated dis- follows,
tribution. The fault conditions are the combinations of six fault
resistances (0 X, 50 X, 100 X, 200 X, 500 X and 1 kX) and 6 FIAs wi ¼ lAi ðxÞ  lBi ðyÞ i ¼ 1; 2: ð14Þ
(Fault Inception Angles) (0°, 18°, 36°, 90°, 126° and 162°). The FIA
is defined as the angle of phase-a. As can be seen, the total number wi ¼ wi =ðw1 þ w2 Þ i ¼ 1; 2: ð15Þ
of fault condition in each fault type is 36. The variances of FIs are

shown in Fig. 3. f1 ¼ p1 x þ q1 y
; ð16Þ
From Fig. 3, these four fault types can be distinguished by com- f2 ¼ p2 x þ q2 y
parison of FIs in 3 steps:
f ¼ w1 f1 þ w2 f2 ; ð17Þ
 AG and ABG faults can be discriminated from AB and ABC/ABCG
faults, because U0 is non-zero from subfigures (a) and (b) in where x and y are the inputs of ANFIS, l is the membership function
Fig. 3 and zero from subfigures (c) and (d). This is because U0 of Ai and Bi which are fuzzy sets associated with the inputs, wi de-
is zero when zero-sequence loops do not exist (AB fault) or sym- notes the firing strength of a rule, wi is the normalized wi, pi and qi
metric fault (ABC/ABCG) occurs. In addition, it has been also are the parameters sets of the consequent parameters, fi is a func-
noted that U0 decreases as the fault resistance increases. tion, and f computes the overall output.
246 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252

1.0 1.0 50

Zero-sequence Voltage U0
Standard deviations (p.u)

Correlation Coefficients
0.9 sa*
sb*
0.8 18°;
0~1k
sc*
0.7 0.9 25
ρa ,b
0.6
ρa ,c
0.5
ρb,c
0.4 0.8 0
1# 15# 30# 36# 1# 15# 30# 36# 1# 15# 30# 36#
Fault Cases Fault Cases Fault Cases
(a)
1.0 1.0

Zero-sequence Voltage U0
Standard deviations (p.u)

Correlation Coefficients
ρa, b 30
0.8 0.8
ρa, c
sa*
0.6 0.6 ρb, c
sb* 20
0.4 sc* 0.4
10
0.2 0.2

0.0 0.0 0
1# 15# 30# 36# 1# 15# 30# 36# 1# 15# 30# 36#
Fault Cases Fault Cases Fault Cases
(b)
1.0 1.0 0.01

Zero-sequence Voltage U0
Correlation Coefficients
Standard deviations (p.u)

0.8 0.8
ρa ,b
0.6
0.6 ρa ,c
0.4 0.4 ρb,c
sa*
0.2 sb* 0.2
sc*
0.0 0.0 0.00
1# 15# 30# 36# 1# 15# 30# 36# 1# 15# 30# 36#
Fault Cases Fault Cases Fault Cases
(c)
1.0 1.0 0.01
Standard deviations (p.u)

Zero-sequence Voltage U0

ρa ,b
Correlation Coefficients

sa*
0.8
sb*
0.8 ρa ,c
0.6 sc* 0.6 ρb, c

0.4 0.4

0.2 0.2

0.0 0.0 0.00

1# 15# 30# 36# 1# 15# 30# 36# 1# 15# 30# 36#
Fault Cases Fault Cases Fault Cases
(d)
Fig. 3. Variance of FIs in different fault types. (a) AG fault, (b) ABG fault (c) ABC/ABCG fault, and (d) AB fault.

In this work, ANFISs are established in MATLAB environment 5. Train and test of ANFISs
[20]. The ANFIS1 has 4 inputs, and ANFIS2 and ANFIS3 both have
six inputs. Each input of ANFIS has two ‘‘Gauss’’ membership func- 5.1. Simulating model
tions. The ANFISs are initialized by grid partition method. From the
knowledge of ANFIS, there are 16 (24) rules in ANFIS1 and 64 (26) For the absence of field data, the simulating model is estab-
rules in ANFIS2 and ANFIS3. Moreover, the ‘‘and’’, ‘‘or’’ and ‘‘defuzz- lished in PSCAD/EMTDC environment to produce fault data [21].
ification’’ methods in ANFISs are selected as ‘‘product’’, ‘‘max’’ and Its configuration is referenced by IEEE-34 workbench. Because
‘‘wtaver’’ respectively. IEEE-34 workbench is neutral directly grounded system [22], some
 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252 247

Vetor C1

ANFIS 1
Vector C2 Vector C3

4 1 2 3 0
ANFIS 2 ANFIS 3
(LLG) (AG) (BG) (CG) (Others)

7 8 9 10
4 5 6
(ABG) (ACG) (BCG) (ABCG/ (AB) (AC ) (BC)
ABC)

over

Fig. 4. Process of fault classification.

sc * s a* sb * U0 modifications are implemented for simulating neutral non-effec-

tively grounded system. The configuration of model is shown in
Fig. 7.
In Fig. 7, E is the external source with impedance Zs =
ANFIS 1
2.3 + j18.4 X. T-1 is transformer and T-2 is Zig-Zag transformer in
substation. The transformers adopt the high frequency model in
[23]. The compensate reactance L = 7.8 H, over compensated
1, 2, 3, 4 or 0
system by 10%. K is the switch which controls the neutral
grounding style of the system. F denotes the fault location and Rf
is the fault resistance. The line parameters are: zero-sequence
ρa,b ρa,c ρb,c sa* sb* sc* impedance z0 = 0.52 + j1.5 X/km, positive- and negative-
impedance z1 = z2 = 0.26 + j0.37 X/km, zero-sequence susceptance
b0 = j0.15  105 s/km, positive- and negative-susceptance b1 = b2
ANFIS 2 = j0.31  105 s/km. The length of lines and the loads are as same
as those in [22].

4, 5 or 6
5.2. Train of ANFISs

To obtain enough data to train ANFISs, ten types of short-circuit

ρa,b ρa,c ρb,c sa* sb* sc* fault with different conditions are assumed in the established
model. The fault conditions are listed in Table 1.
All fault cases are simulated in the isolated system (with switch
ANFIS 3 K open). The obtained data is used to train ANFISs in MATLAB envi-
ronment. In train step, the stop criteria are that the train epochs
reaches 10,000 or the RMSE (Root Mean Square Error) is smaller
7, 8, 9 or 10 than 0.0003. The convergence curves of ANFISs after train are
shown in Fig. 8.
Fig. 5. Inputs and outputs of ANFISs. From Fig. 8, the train of ANFIS1 and ANFIS3 is completed at
8328 and 703 epochs respectively, since the minimum RMSE is
reached. The train of ANFIS2 is stopped when the maximum
A1 x y epochs are reached, and the final RMSE is about 0.0067.
x w1 w1
A2 II N f1 w 1 f1
5.3. Test of ANFISs

B1 x y Σ f To validate the classification accuracy of trained ANFISs, 3600

y w2 w2 w2 f2 fault cases which are different from those in train step are simu-
B2 II N f2
lated. Their fault conditions are listed in Table 2.
As can be seen in Table 2, the maximum fault resistance
(1.5 kX) is more than half of that in train. It is considered to be
layer 1 layer 2 layer 3 layer 4 layer 5
difficult for ANFISs, because the extrapolation ability of neural
Fig. 6. General structure of ANFIS. networks is not as good as its interpolation ability.
248 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252

Fig. 7. Simulating model in PSCAD/EMTDC.

Table 1
100 100 100 100 100 100 100 100
Different fault conditions in train step. 100.0
Faulted bus 802, 810, 814, 822, 826, 856, 864, 844, 838, 840 99.9
Fault resistance (X) 0, 50, 100, 200, 500, 1000 99.8
99.8

Accuracy Rate (%)

FIA (°) 0, 18, 36, 90, 126, 162 99.7
Total number 3600 99.7
99.6
99.5
99.4 99.4
0.14 99.4
ANFIS1
99.3
ANFIS2
ANFIS1 99.2
ANFIS3
ANFIS2 99.1
ANFIS3
99.0
AG CG Others ACG ABC/ AB BC
BG LLG ABG BCG ABCG AC
RMSE

Fig. 9. Statistic results of test cases.

The approach proposed in this paper

The approach proposed in [10]
100

95
0.00
0 10000
Accuracy Rate (%)

Epoches 90

Fig. 8. Convergence curves of ANFISs after train. 85

80
Table 2
Different fault conditions in test step. 75

Faulted buses Fault resistance (X) FIA (°) Number 70

806, 816, 818, 830 0, 5, 10, 25, 40, 50 0, 18, 54, 72, 108, 144 1440
812 0, 50, 100, 200, 500, 1000 0, 18, 36, 90, 126, 162 360 65
832, 834 20, 70, 150, 300, 600, 1200 18, 54, 72, 90, 108, 144 720 AG CG ACG ABC/ AB BC
848, 860, 862 20, 70, 150, 300, 1200, 1500 18, 54, 72, 90, 108, 144 1080 BG ABG BCG ABCG AC
Total 3600
Fig. 10. Comparison of two approaches.

From Fig. 9, the minimum d is 99.4% in ABG faults which means 2

All fault cases in Table 2 are simulated. The obtained data is
out of 360 ABG faults are misclassified. In addition, through our
used to test ANFISs. Because the outputs of ANFIS are not always
observation, the classification results are right when fault resistance
integers, the round off method is adopted here. From the test
is 1.2 kX and 1.5 kX. It reveals that the trained ANFISs have good
results, the accuracy of ANFISs is very good. Due to the space
extrapolation ability.
constraint, it is impossible to list all results in the paper. Hence,
In comparison with the other proposed approaches [8–11], this
the statistic results are shown in Fig. 9.
work needs high sampling devices and complicated calculation. If
The accuracy rate d in Fig. 9 is defined by,
it does not have advantages, it would not be practical. So, the pro-
Right numbers from ANFIS classification posed approach is compared with the approach in the paper [10]
d¼  100% ð18Þ
360ðTotal numbers in each fault typeÞ for its transient-based either and recent publication. This approach
 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252 249

150 2
-Phase C

-Phase B

-Phase A
50
Current i (A)

Voltage (kV)
0
0

-50

-150
0.1 0.2 0.3
-2
Time t (s) -160 0 160
(a) Current (A)

Fig. 12. V–I character of arc model.

0.4
-Phase C

-Phase B
0.2 100
-Phase A 100.0
ANFIS1
Current i (A)

99.8
0 99.6
Accuracy Rate (%)
99.4
99.2
-0.2
99.0

98.8
-0.4 98.6
98.6
31 210
Sampling Points
98.4 98.3
(b) 98.2
Fig. 11. Currents of three phases, FIA = 70°, fault resistance is 70 X, (a) the original, 98.0
and (b) the WT-extracted. AG BG CG

Fig. 13. Statistic results under arc faults.

extracts the energies of transient currents and compares them with

the thresholds to obtain the fault classification result. There are
three thresholds which are k3 (the minimum classification index),
k4 (the minimum relative classification index) and k5 (the minimum
zero-sequence current relation). They equal 2000, 2 and 30 respec-
tively in our work. The fault cases listed in Table 2 are used to com-
pare two approaches, and the results are shown in Fig. 10.
As can be seen from Fig. 10, the approach proposed in this paper
exhibits dominant performance. For the approach proposed in the
paper [10], it is difficult to find appropriate thresholds which are
adaptable to different fault conditions. Take ACG fault at 848 node
of the simulating model for example, the original currents and
WT-extracted currents are shown in Fig. 11.
Through calculation from (b) in Fig. 11, the energies of three
phases are Ea = 1.89, Eb = 1.9 and Ec = 1.57. Since the energy of
healthy phase Eb is even bigger than those of faulty phases Ea
and Ec, the approach is confused. However, the approach proposed
in this paper is right in such fault case.
6. Adaptability analysis
6.1. Neutral compensated grounding distribution
Let the switch K close in Fig. 7, then the distribution model
becomes neutral compensated grounding. The fault cases in
Table 2 are simulated again and the obtained data is used to verify
the proposed approach.
Although the proposed approach is trained in the isolated net-
work, the results show that it is hardly affected by the change of
neutral grounding style. The statistic results are similar to those
in Fig. 9, so they are not repeated here. As we know, the response
of suppress coil in high frequency band is quite slow in comparison
with that of the distributed capacitances along feeder. So, the WT-
extracted current signals are quite same as those in the isolated
distribution during the first post-fault cycle. It is intuitively why
the test results are quite same as those in the isolated distribution.
6.2. Non-linear fault resistance
In practice, fault always accompanies arc phenomenon. The
fault resistance of arc is non-linear and varies with wind speed,
environmental temperature and so on. For study on the impact
of the non-linear fault resistance, The arc model and its extinction
condition introduced in the paper [24] are employed in this work.
There are three parameters to initialize arc model, which are s0
(the initial time constant), l0 (the initial length) and a (the negative
exponent). They equal 0.25 ms, 0.2 m and 0.4 respectively. Those
values are considered to be typical in power distribution network.
The simulating V–I character of arc model is presented in Fig. 12.
In comparison with the result reported in the paper [25], Fig. 12
shows that the established arc model is very effective. Because SLG
fault parameters are only provided in the paper [24], 1080
(360  3) SLG test cases contained in Table 2 are simulated with
250 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252
the arc fault. The accuracy of proposed approach to identify SLG
(Single-Line-to-Ground) faults is verified, and the statistic results
are shown in Fig. 13.
Because ANFIS1 actualizes the SLG faults classification, the
accuracy rate of ANFIS1 is only shown in Fig. 13. In comparison
with Fig. 9, Fig. 13 shows that the non-linear fault resistance does
not decrease the accuracy rate monotonically since it is enhanced
in CG faults. In AG and BG faults, the accuracy rate decreases
slightly. In BG fault, it reaches 98.3% which means 6 of 360 cases
are misclassified. The result is acceptable.
In conclusion, the non-linear fault resistance has little impact
on the proposed approach in SLG faults.
Fig. 14. Different network configurations.
100100 100 100100100100
100 99.7 99.8
99.5 99.699.699.6
99
Accuracy Rate (%)
98
97
96.1 ANFIS1
96 ANFIS2
ANFIS3
95
94.3
94
O a b c d O a b c d O a b c d
Fig. 15. Statistic results under different network configurations.
6.3. Different network configurations
Switch operations are common in distribution network to actu-
alize load balance, power loss minimum and other purposes. So,
the configuration of distribution network is not constant. The fault
classification approach must show good performance regardless of
different network configurations. For the purpose, four different
network configurations shown in Fig. 14 are considered to verify
the proposed approach.
As can be seen from Fig. 14, the configurations are obtained
from the original in Fig. 7, i.e. some laterals of the original config-
uration are deleted to form the new configurations. The toughest
case is (d) in Fig. 14, because the 86% line length of total is deleted
and only 5 nodes (800, 802, 806, 808 and 810) are left.
For comparison with the results in different configurations, ten
types of short-circuit fault are assumed at 806 node, and the fault
conditions are the combinations of 10 fault resistances (0 X, 50 X,
100 X, 200 X, 400 X and 700 X) and 10 FIAs (0°, 18°, 36°, 90°, 126°
and 162°). The obtained data is used to verify the proposed ap-
proach. The statistic results are shown in Fig. 15.
In Fig. 15, the symbol ‘O’ on x-axis denotes the original config-
uration, and the symbols ‘a’, ‘b’, ‘c’ and ‘d’ correspond to the config-
urations in Fig. 14.
As can be seen from Fig. 15, the more laterals are deleted, the
more the accuracy rate decreases. In (a) and (b) configurations,
there is no obvious decrease of the statistic results. In (c) and (d)
configurations which are 41% and 14% of the total lines left in
the original network, the accuracy rate decreases. The minimum
accuracy rate is 94.3% when ANFIS2 classifies LLG faults in (d) con-
figuration. The results are acceptable, because they are above 90%.
6.4. Different load levels
To study the impact of different load levels on the proposed ap-
proach, two load levels are considered which are ‘‘half loading’’ and
‘‘double loading’’. The term ‘‘half loading’’ indicates that the loads
at all buses are reduced by 50% uniformly, and the term ‘‘double
loading’’ indicates that the loads at all buses are doubled uni-
formly. The test cases in Table 2 are simulated again under each
load level. The obtained data is used to verify the proposed ap-
proach, and the statistic results are shown in Fig. 16.
In comparison with Fig. 9, the subfigure (a) in Fig. 16 shows that
the decrease of load has little influence on the proposed approach.
Moreover, the accuracy is benefit from the light load level, because
the accuracy rate increases in CG faults and ABG faults. In compar-
ison with Fig. 9, the subfigure (b) in Fig. 16 shows that the heavy
load decreases the accuracy rate of proposed approach remarkably,
especially in SLG faults. The minimum accuracy rate is 83.3% when
 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252 251

100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100
100.0 100.0
99.9 99.9
99.9

Accuracy Rate (%)

99.8
Accuracy Rate (%)

99.7 99.7 99.8

99.7
99.6
99.6 99.7
99.7
99.5
99.5
99.4 99.6
ANFIS1 ANFIS1
99.3 99.5
ANFIS2 99.5 ANFIS2
99.2 99.4
ANFIS3 99.4 ANFIS3
99.1
99.0 99.3
AG CG Others ACG ABC/ AB BC AG CG Others ACG ABC/ AB BC
BG LLG ABG BCG ABCG AC BG LLG ABG BCG ABCG AC
(a) (a)
100 99.7 100 99.7 99.6 100 99.5 100 100 100 100 100 100 100 100 100
100 99.3 100.0
98
99.9
96

Accuracy Rate (%)

Accuracy Rate (%)

94 99.8
92 99.7 99.7
99.7
90 89.2 88.9
88 99.6
ANFIS1
86 99.5 ANFIS1
ANFIS2 99.5
84 83.3 ANFIS2
ANFIS3 99.4
99.4 ANFIS3
82
80 99.3
AG CG Others ACG ABC/ AB BC AG CG Others ACG ABC/ AB BC
BG LLG ABG BCG ABCG AC BG LLG ABG BCG ABCG AC

(b) (b)
Fig. 16. Statistic results under different load levels. (a) half loading, and (b) double Fig. 17. Statistic results under different source impedances. (a) condition No. 1, and
loading. (b) condition No. 2.

Table 3 of source impedance has very negligible impact on the proposed

Source impedance conditions undertaken for study.
approach. The accuracy rate obtained under condition Nos. 1 and
No. Impedance conditions 2 is shown in Fig. 17.
1 SE + ve seq. impedance magnitude is reduced by 70%, rest unchanged Comparison of Fig. 17 with Fig. 9, their statistic results are quite
2 SE + ve seq. impedance magnitude is reduced by 70%, rest unchanged same. The results obtained under condition Nos. 3, 4 and 5 are as
3 SE + ve seq. impedance angle is made 20°, rest unchanged same as those shown in Fig. 17, so they are not repeated here.
4 SE + ve seq. impedance angle is made 50°, rest unchanged
5 SE + ve seq. impedance magnitude is doubled, rest unchanged
7. Conclusion

In this paper, wavelet technique is employed to extract the

ANFIS1 identifies CG faults. It means 60 out of 360 faults are mis-
transient signals in the interested frequency band. The statistic
classified. The result is unacceptable. Fortunately, faults are in-
quantities are used to construct FIs. An ANFIS-based fault classifi-
clined to happen when the load are heavy in the field, and this
cation approach in distribution network is proposed. Through a
means much fault data under heavy load can be obtained in history
large number of simulation studies, the main conclusions of this
and used to train ANFISs. Since the train samples contain a lot of
work are given as follows:
those in heavy load level, the performance of ANFIS can be highly
improved.
 The constructed FIs can effectively characterize the fault type.
 The proposed approach only needs three phase currents and
6.5. Variance of source impedance zero-sequence voltage. It can accurately classify ten types of
short-circuit faults in distribution. Moreover, it can actualize
The source impedance depends on the system condition at the high resistance fault classification. The highest resistance tested
SE (Substation End). To study the impact of source impedance on in this paper is 1.5 kX.
the proposed approach, the impedance at SE has been varied as  Although the ANFISs are trained in the isolated distribution, it
outlined in Table 3 [8]. can be adaptable to the neutral compensated grounding distri-
Under each of impedances given in Table 3, the test cases listed bution without any change. Besides, the variances of source
in Table 2 are simulated. Comparison of the results thus obtained impedance and fault resistance (arc fault) have little impact
with those obtained with constant source impedance, the variation on the proposed approach.
252 J. Zhang et al. / Electrical Power and Energy Systems 49 (2013) 243–252
 When the change of network configurations is not intensive, the
accuracy rate of proposed approach is as same as that obtained
in the original configuration. As the network configuration
changes more intensively, the accuracy rate decreases more
obviously. However, the accuracy rate keeps above 90% in our
work.
 With regard to the impact of load level, the light load has little
impact on the proposed approach, but the heavy load decreases
the accuracy rate remarkably. Fortunately, faults tend to hap-
pen when load is heavy. A great deal of history fault data in
heavy load can be used to train ANFISs and improves their
accuracy.
In our future work, the performance of proposed approach
should be improved through increasing the train samples of heavy
load from the field. The history fault data in power system will be
collected, and used to verify the proposed approach.
References
[1] Upendar J, Gupta CP, Singh GK. Statistical decision-tree based fault
classification scheme for protection of power transmission lines. Int. J. Electr.
Power Energy Syst. 2012;36(1):1–12.
[2] Valsan SP, Swarup KS. High-speed fault classification in power lines: theory
and FPGA-based implementation. IEEE Trans. Ind. Electron.
2009;56(5):1793–800.
[3] Safty SE, El-Zonkoly A. Applying wavelet entropy principle in fault
classification. Int. J. Electr. Power Energy Syst. 2009;31(10):604–7.
[4] El-Zonkoly AM, Desouki H. Wavelet entropy based algorithm for fault
detection and classification in FACTS compensated transmission line. Int. J.
Electr. Power Energy Syst. 2011;33(8):1368–74.
[5] Parikha UB, Dasb B, Maheshwarib R. Fault classification technique for series
compensated transmission line using support vector machine. Int. J. Electr.
Power Energy Syst. 2010;32(6):629–36.
[6] Bhowmika PS, Purkaitb P, Bhattacharyac K. A novel wavelet transform aided
neural network based transmission line fault analysis method. Int. J. Electr.
Power Energy Syst. 2009;31(5):213–9.
[7] D.Q. Hou, N. Fischer, Deterministic High-Impedance Fault Detection and Phase
Selection on Ungrounded Distribution Systems, in: IEEE/IAS Ind. & Com. Pow.,
Edmonton, vol. 2006, pp. 112–122.
[8] Das B. Fuzzy logic-based fault-type identification in unbalanced radial power
distribution system. IEEE Trans. Power Del. 2006;21(1):278–85.
[9] Thukaram D, Khincha HP, Vijaynarasimha HP. Artificial neural network and
support vector machine approach for locating faults in radial distribution
systems. IEEE Trans. Power Del. 2005;20(2):710–21.
[10] Salim RH, Oliveira Kd, Filomena AD, Resener M, Bretas AS. Hybrid fault
diagnosis scheme implementation for power distribution systems automation.
IEEE Trans. Power Del 2008;23(4):1846–56.
[11] Dwivedi UD, Singh SN, Srivastava SC. A wavelet based approach for
classification and location of faults in distribution systems. Annual IEEE
India, vol. 2. KANKUR; 2008. p. 488–93.
[12] Dong XZ, Shi SX. Identifying single-phase-to-ground fault feeder in neutral
noneffectively grounded distribution system using wavelet transform. IEEE
Trans. Power Del. 2008;23(4):1829–37.
[13] D. Stefanoiu, F. Lonescu, A fuzzy-statistical reasoning model for bearings fault
diagnosis, in: IEEE Systems, Systems Man and Cybernetics, vol. 7, 2002, pp. 6–
11.
[14] Erenturk K. ANFIS-based compensation algorithm for current-transformer
saturation effects. IEEE Trans. Power Del. 2009;24(1):195–201.
[15] Reddy MJ, Mohanta DK. Adaptive-neuro-fuzzy inference system approach for
transmission line fault classification and location incorporating effects of
power swings. IET Gener. Transm. Dis. 2008;2(2):235–44.
[16] Etemadi AH, Sanaye-Pasand M. High-impedance fault detection using multi-
resolution signal decomposition and adaptive neural fuzzy inference system.
IET Gener. Transm. Dis. 2008;2(1):110–8.
[17] Reddy M, Mohanta DK. Performance evaluation of an adaptive-network-based
fuzzy inference system approach for location of faults on transmission lines
using Monte Carlo simulation. IEEE Trans. Fuzzy Syst. 2008;16(4):909–19.
[18] Zhang Y, Zhou Q, Sun CX, Lei SL, Liu YM, Song Y. RBF neural network and
ANFIS-based short-term load forecasting approach in real-time price
environment. IEEE Trans. Power Syst. 2008;23(3):853–8.
[19] Makarand SB, Zafar JK, Hiralal MS, Ram LS. Adaptive neural fuzzy inference
system for the detection of inter-turn insulation and bearing wear faults in
induction motor. IEEE Trans. Ind. Electron. 2007;54(4):250–8.
[20] User’s Guide (MATLABÒ Compiler™), second ed., Math Works Inc., Natick,
2000.
[21] PSCAD/EMTDC User’s manual, Manitoba HVDC Research Centre Inc.,
Winnipeg, 2003.
[22] W.H. Kersting, Radial distribution test feeders, in: IEEE Power Engineering
Society Winter Meeting, Columbus, vol. 2, 2001, pp. 908–912.
[23] Borghetti A, Bosetti M, Silvestro MD, Nucc CA, Paolone M. Continuous-wavelet
transform for fault location in distribution power networks: definition of
mother wavelets inferred from fault originated transients. IEEE Trans. Power
syst. 2008;23(2):380–8.
[24] M. Kizilcay, P. Seta, Digital simulation of fault arcs in medium-voltage
distribution networks, in: PSCC 15th Power Systems Computation, Liege, vol.
2005, pp. 22–26.
[25] Michalik M, Rebizant W, Lukowicz M, Lee SJ, Kang SH. High-impedance fault
detection in distribution networks with use of wavelet-based algorithm. IEEE
Trans. Power Del. 2006;21(4):1793–802.