Contenu connexe
Similaire à Ann and impedance combined method for fault location in electrical power distribution systems (20)
Plus de IAEME Publication (20)
Ann and impedance combined method for fault location in electrical power distribution systems
- 1. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
INTERNATIONAL JOURNAL OF ELECTRICAL ENGINEERING &
TECHNOLOGY (IJEET)
ISSN 0976 – 6545(Print)
ISSN 0976 – 6553(Online)
Volume 5, Issue 9, September (2014), pp. 29-38
© IAEME: www.iaeme.com/IJEET.asp
Journal Impact Factor (2014): 6.8310 (Calculated by GISI)
www.jifactor.com
IJEET
© I A E M E
ANN AND IMPEDANCE COMBINED METHOD FOR FAULT LOCATION IN
ELECTRICAL POWER DISTRIBUTION SYSTEMS
Zahri Mustapha1, Menchafou Youssef2, El Markhi Hassane2, Habibi Mohamed1
1Laboratory of Telecommunication and Engineering Decision Systems, Kenitra, Morocco
2Laboratory of Signals, Systems and components Fez, Morocco
29
ABSTRACT
Power distribution systems play important roles in modern society. When distribution system
outages occur, fast and proper restoration are crucial to improve the quality of services and customer
satisfaction. In this paper, an accurate algorithm is suggested for single line to ground fault location.
In this algorithm we use the parameters (fault resistance and apparent impedance) as inputs of an
artificial neural network; (ANN) has been chosen to give in output the faulty section. Test results are
obtained from numerical simulation using a real under-ground distribution feeder data from “Inde-pendent
agency for the distribution of water and Electricity, and liquid cleansing of Kenitra”,
Morocco.
Keywords: Artificial Neural Network, Apparent Impedance, Fault Location, Fault Resistance,
Single Line-To-Ground Fault.
1. INTRODUCTION
Distribution networks are dispersed in each urban and rural region, and are crossed from each
alley and street. Each distribution feeder has many laterals, sub-laterals, load taps, balanced and unba-lanced
load and different types of conductors [1]. Power distribution systems (PDSs) are subjected to
fault conditions caused by various sources such as adverse weather conditions, equipments failure
and external object contacts. Owing to the expansion of distribution network, it is very difficult and
complicated to locate the fault in these networks.
Currently the only technique used for locating faults in distribution systems of electric power
is the visual inspection of FPI (Fault Passage Indicators) which imposes an important time of restora-tion.
In recent years, many techniques have been proposed for automated FL (Fault Location) in the
distribution systems [2-6], but these methods aim the location of the exact position of the fault which
- 2. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
affects their accuracy, and hinders their practical implementation. This led us to work on the location
of the faulty section knowing that the companies already have the tools to search for the exact posi-tion
of the fault in the section located off (vehicle of fault research). In order to efficiently increase
the level of automation, and reduce the time of restoration of distribution systems, a new algorithm is
presented in this article Fig. 1.
Figure. 1: The proposed algorithm.
The main objective of the proposed methodology is to determinate apparent reactance which
varies with the fault resistance; these parameters (RF, Xm) are used as inputs of an artificial neural
network (ANN) to have in the output the faulty section.
The proposed method was validated using real under-ground distribution feeder data and Mat-lab
30
as an analysis tool.
The remainder of this paper is organized as follow: The estimation of the apparent reactance
and the fault resistance is explained in sections 2 and 3 respectively; Section 4 presents the use of
ANN to locate the faulty section of line; the tests’ results are shown in Section 5, whereas the conclu-sions
of this work are presented in Section 6.
- 3. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
2. ESTIMATING THE APPARENT REACTANCE
The equations needed to calculate the apparent impedance and reactance from node t to the
fault depend on the type of the fault. For us we use the phase to ground fault parameters using the
voltage and current at the time of fault and neglecting the effects of varying loads [7].
am
Z = (1)
am
−
I I
1
F m Fmi m Fmr V
31
V
m I
Im( ) m m X = Z
Where
Zm is the apparent impedance from node t to the fault;
Xm is the apparent reactance from node t to the fault;
Vam is phase m sending-end voltage (in Volts);
Iam is phase m sending-end current (in amperes).
3. FAULT RESISTANCE ESTIMATION
After detection and classification of the faults, the FL process is initialized. First, the system is
divided into n branches, where n is the number of possible paths (end nodes). For each branch, the
fault resistance is estimated, using an impedance-based method [8-12]. After the estimation of the
fault resistance and the apparent reactance, the procedure for determining the faulty section is started.
Figure. 2: Single-phase-to-ground fault modelling
3.1 Mathematical development
The system illustrated in Fig. 2, contains a local bus, a generic faulted distribution line with
constant fault resistance (RF), and an equivalent load.
It is possible to show that for a single phase-to-ground fault in phase m:
−
−
=
Smr
Smi
Fmi Fmr
m m
V
M M
x
R M I M I
.
. .
1 2 2 1
(2)
Where the subscript indices r and i represent, respectively, the variables real and imaginary
parts; the variables are as follow:
Vsm phase m sending –end voltages (in volts);
VFm phase m fault-point phase voltages (in volts);
x fault point to local bus distance (in kilometers);
IFm fault current (in amperes).
- 4. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
m mkr Skr mki Ski M (Z I Z I ) 1 (3)
m mkr Ski mki Skr M (Z I Z I ) 2 (4)
− +
. .
M V M V
2 1
m Smr m Smi
= (5)
F M I −
M I
Z Z Z
aa ab ac
Z Z Z
ba bb bc
32
Also, M1m and M2m are defined in (3) and (4)
= −
k
= −
k
Where
k phases a,b and c;
Zmk impedance between phase m and k [/km];
ISk phase k sending-end current (in amperes).
The fault resistance is estimated by (5):
m Fmi m Fmr
R
. .
1 2
From (5) it is possible to obtain the fault resistance from the parameters of the system: the
fault current and the sending-end voltages. Since voltages are already known, an iterative procedure
that updates the fault current is used to estimate the fault resistance.
3.2 Fault Current Estimation Procedure
In equation (5) the only unknown parameter is the fault current IFmr,i . All other variables are
system parameters or measured variables.
Referring to Fig. 2, the fault current can be obtained by (6)
Fa Sa Ra Sa La I = I + I = I − I (6)
Where ILa is the phase a load current.
Nevertheless, the load current during the fault period is different from the prefault load cur-rent,
due to voltage drops and systems dynamics during the fault. For this reason, an iterative tech-nique
used to estimate the load current during the fault, is described as follow: [12]
1) Load current during the fault ILa is assumed to be the same as the prefault load current.
2) The fault current is calculated using (6)
3) Fault resistance is estimated using (2), (3), and (4).
4) Fault-point voltages are estimated using (7)
−
=
V
V
V
V
Fa
Fb
Fc
ca cb cc
Sa
Sb
Sc
Fa
Fb
Fc
I
I
I
Z Z Z
x
V
V
. (7)
5) Load current ILa is updated using the fault-point voltages in (7) and (8):
[ ][ ]T
La aa ab ac Fa Fb Fc I = Y Y Y .V V V (8)
- 5. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
[ ( ) ] 1
− = − + pq pq Lpq Y l x Z Z (9)
33
Where
Zpq is the line impedance (mutual or self) between phase p and q;
ZLpq is the load impedance (mutual or self) between phase p and q;
And l is the total line length.
6) Check if RF has converged, using (10)
(a ) − (a −1) d F F R R (10)
Where is a previously defined threshold value and is the iteration number.
7) If RF has converged, stop the procedure; otherwise, go back to step2).
The output of this procedure is the fault resistance, used with apparent reactance as inputs of
our neural network.
4. NEURAL NETWORK STEP
4.1 Neural Network Application to FL
Artificial Neural Networks (ANNs) can be applied to fault analysis because they are a
programming technique applicable to problems in which the information appears in a vague,
redundant, distorted, or massive form. Also, they are able to learn using examples [13].
In the problems of FL, they are potentially applicable because:
• Many parameters must be considered;
• There are methods to simulate examples in a quick and reliable way;
• The ANN output is very fast, because its working consists in a series of very simple operations.
Although the programming using ANNs has great advantages, it also presents some disadvan-tages.
Among them, the complexity of the type and network architecture selection (number of layer,
number of neurons per layer, activation functions, learning algorithms parameters, etc.) can be em-phasized.
The FL problem in distribution lines using ANNs consists in defining a neural network that
allows obtaining the faulty section, using a small number of electrical parameters measured in a line
terminal.
4.2 Selecting the Right Architecture
One factor in determining the right size and structure for the network is the number of inputs
and outputs that it must have. The lower the number of inputs, the smaller the network can be. How-ever,
sufficient input data to characterize the problem must be ensured. As the apparent reactance
includes the fault information in spite of varying with the fault resistance, we have chosen to use it as
inputs of our ANN to have the faulty section in the output. Thus the network inputs X and output Y
is:
X = [ ] m F X , R (11)
- 6. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
Y= [number of the faulty section] (12)
Once how many inputs and outputs the network should have was decided, the number of layer
and the number of neurons per layer was considered. After several trials, it was decided to use a neur-al
network with thirty neurons in the hidden layer.
The final determination of the neural network requires the relevant transfer functions to be
established. After analyzing various possible combinations of transfer functions used, such as logsig,
tansig and linear functions, the tansig function was chosen as transfer function for the hidden layer,
and the purelin function in the output layer [14].
34
4.3 Learning Rule Selection
The back-propagation learning rule is used in perhaps 80-90% of practical applications [15].
Improvement techniques can be used to make back-propagation more reliable and faster. The back-propagation
learning rule can be used to adjust the weights and biases of networks to minimize the
sum-squared error of the network. As the simple back-propagation method is slow because it requires
small learning rates for stable learning, improvement techniques such as momentum and adaptive
learning rate or alternative method to gradient descent, Levernberg-Marquard optimization can be
used. Various techniques were applied to the different network cases, and it was concluded that the
most suitable training method for architecture selected was based on the Levernberg-Marquard opti-mization
technique.
4.4 Training Process
Order to train the network, a suitable number of representative examples of relevant pheno-menon
must be selected so that the network can learn the fundamental characteristics of the problem,
and once the training is completed, the network provides correct results in new situations not envi-saged
during training.
The system that we have studied is a part of the underground distribution network of Indepen-dent
Agency for the distribution of water and Electricity and liquid cleansing of Kenitra, Morocco.
It is a line from 20 KV distribution network of total length 7.5 Km, composed of 18 sections
of different lengths, simulated using distributed parameter line model as shown in Table 1.
Matlab [16] simulates several cases of fault at different FL between 0-100% and for several
fault resistances 0-100. Preprocessing is used to reduce the size of the neural network and improve
the performance and the speed of learning. [17]
The apparent reactance and the fault resistance are calculated for each case using the sending-end
current and voltages; the inputs were normalized in order to limit them between -1 and +1; the
number of simulated faults is 10 at different locations and for 10 different resistances. Therefore, the
total number is 18 * 10 * 10 = 1800.
The next step is to divide the data up into 70% for training, 20% for validation and 10% for
test. The mean square error of our network is shown in Fig. 3.
5. TEST RESULTS
After training, the proposed algorithm is extensively tested using data never envisaged to
verify the effect of the fault location (close or far from the input of the section), and of the fault resis-tance
(0-100 ) on the performance of this method.
The performances of a neural network are usually measured by the errors on the training,
validation and test sets, but often useful to investigate the network response in more detail.
Fig. 4 shows four curves, the first shows the outputs of training data according to their desired
outputs, the second track outputs of validation data according to their desired outputs, the third
- 7. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
shows the outputs of test data according to their desired outputs and the fourth groups all the curves
presented before. We can easily see that the outputs obtained have values around the desired value so
this is a good result in condition not to have two confused outputs or an output far than their desired
one. To verify this condition, the mean square error must have values between -0.5 and +0.5 to
approach it to the desired value Fig. 5. Or values between -1 and +1 if we use MATLAB function
round to close all numbers to the nearest integer Fig. 6.
In analyzing the results we can see that 98.5% of the values of the curve are located within the
range of tolerance that has already defined [-0.5 0.5], which lets say 98.5 % of faults are recognized.
Figure. 3: Training figure using Levenberg-marquardt algorithm
Figure. 4: Regression analysis of training, validation and test data
35
- 8. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
Figure. 5: Error on the outputs
Figure. 6: Error on the outputs using MATLAB function ROUND
TABLE 1: STUDIED SECTIONS OF LINES PARAMETERS
Section of
36
line
Length (Km)
Resistance
(/km)
Reactance
(/Km)
1 2.5 0.161 0.155
2 0.268 0.161 0.155
3 0.175 0.161 0.155
4 0.195 0.161 0.155
5 0.840 0.2390 0.110
6 0.212 0.2390 0.110
7 0.306 0.2390 0.110
8 0.193 0.2390 0.110
9 0.233 0.2390 0.110
10 0.241 0.2390 0.110
11 0.168 0.2390 0.110
12 0.241 0.2390 0.110
13 0.288 0.2390 0.110
14 0.155 0.2390 0.110
15 0.407 0.2390 0.110
16 0.118 0.2390 0.110
17 0.290 0.2390 0.110
18 0.680 0.2390 0.110
- 9. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
37
6. CONCLUSION
The proposed algorithm allows us to locate the faulty section using the neural network that
receives the apparent reactance and resistance of fault, calculated using the sending-end current and
voltage, as inputs.
The performances of this algorithm are verified by several tests simulating 1800 case of phase
to ground faults.
Simulation results show that 98.5% of simulated faults are exactly located, and for the rest
(1.44%), this method indicates that the fault is located in the section before or after the one that is
really faulty.
7. REFERENCES
[1] R. Dashti, J. Sadeh, “Accuracy improvement of impedance-based fault location method for
power distribution network using distributed-parameter line model,” Euro. Trans. Electr.
Power, DOI:10.1002/etep/1690, no. 17, Sep.2012.
[2] S.-J. Lee, M.-S. Choi, S.-H. Kang, B.-G. Jin, D.-S. Lee, B.-S. Ahn, N.-S. Yoon, H.-Y. Kim,
and S.-B. Wee, “An intelligent and efficient fault location and diagnosis scheme for radial
distribution systems,” IEEE Trans. Power Del., vol. 19, no. 2, pp. 524–532, Apr. 2004.
[3] M.-S. Choi, S.-J. Lee, D.-S. Lee, and B.-G. Jin, “A new fault location algorithm using direct
circuit analysis for distribution systems,” IEEE Trans. Power Del., vol. 19, no. 1, pp. 35–41,
Jan. 2004.
[4] J. Zhu, D. Lubkeman, and A. Girgis, “Automated fault location and diagnosis on electric
power distribution feeders,” IEEE Trans. Power Del., vol. 12, no. 2, pp. 801–809, Apr. 1997.
[5] E. Senger, J. Manassero, G. , C. Goldemberg, and E. Pellini, “Automated fault location
system for primary distribution networks,” IEEE Trans. Power Del., vol. 20, no. 2, pt. 2,
pp. 1332–1340, Apr. 2005.
[6] André D. Filomena, Mariana Resener, Rodrigo H. Salim, Arturo S. Bretas. “Fault location for
underground distribution feeders: An extended impedance-based formulation with capacitive
current compensation”. Electr Power and Energy Syst 2009; 31:489–496.
[7] R. Das, “Determining the location of faults in distribution systems”, Ph.D. dissertation, Univ.
Saskatchewan, Saskatoon, SK, Canada, 1998.
[8] R. Hartstein Salim, M. Resener, A. Darós Filomena, K. Rezende Caino d’Oliveira, A. Suman
Bretas, “Extended Fault-Location Formulation for Power Distribution Systems”, IEEE
TRANSACTIONS ON POWER DELIVERY, VOL. 24, NO. 2, APRIL 2009.
[9] R. Hartstein Salim, M. Resener, A. Darós Filomena, K. Rezende Caino d’Oliveira,
A. Suman Bretas, “Hybrid Fault Diagnosis Scheme Implementation for Power Distribution
Systems Automation”, IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 4,
OCTOBER 2008.
[10] A. Suman Bretas,R. Hartstein Salim, “Fault Location in Unbalanced DG Systems using the
Positive Sequence Apparent Impedance”, IEEE PES Transmission and Distribution Confe-rence
and Exposition Latin America, Venezuela, 2006.
[11] R. Hartstein Salim, M. Resener, A. Darós Filomena, A. Suman Bretas, “Ground Distance
Relaying With Fault-Resistance Compensation for Unbalanced Systems”, IEEE
TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 3, July 2008.
[12] André D. Filomena, Mariana Resener, Rodrigo H. Salim, Arturo S. Bretas. Distribution
systems fault analysis considering fault resistance estimation. Electr Power and Energy Syst
2011; 33:1326–1335
- 10. International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 – 6545(Print),
ISSN 0976 – 6553(Online) Volume 5, Issue 9, September (2014), pp. 29-38 © IAEME
[13] J. Gracia, A.-J. Mazon, and I. Zamora, “Best ANN Structures for Fault Location in Single-and
Double-Circuit Transmission Lines,” IEEE Trans. Power Del., vol. 20, no. 4, Oct. 2005.
[14] T. Dalstain, and B. Kulicke, “Neural network-approach to fault classification for high speed
protective relaying”, IEEE Trans. On Power Delivery, vol. 10, No. 2, Apr. 1995,
pp. 1002-1011.
[15] Anamika Jain, A.S. Thoke, and R.N. Patel, “Classification of Single Line to Ground Faults on
Double Circuit Transmission Line using ANN”, International Journal of Electrical Systems
Science and Engineering, vol. 1, No. 2, June 2009, pp. 1793-8163.
[16] R. Patel and K.V. Pagalthivarthi, “MATLAB-based modelling of power system components
in transient stability analysis”, International Journal of Modelling and Simulation, vol. 25,
No. 1, 2005, pp. 43-50.
[17] H. Demuth M Beale, Neural Network –toolbox – for Use with Matlab®. Available:
38
http://www.mathworks.com.
[18] Soumyadip Jana, Sudipta Nath and Aritra Dasgupta, “Transmission Line Fault Classification
Based on Wavelet Entropy and Neural Network”, International Journal of Electrical
Engineering Technology (IJEET), Volume 3, Issue 2, 2012, pp. 94 - 102, ISSN Print:
0976-6545, ISSN Online: 0976-6553.
[19] Santosh B. Kulkarni, Rajan H. Chile, Azhar Ahmed and Arvind R. Singh, “Performance
Evaluation of Fault Location Algorithm for Protection of Series Compensated Transmission
Line”, International Journal of Electrical Engineering Technology (IJEET), Volume 2,
Issue 2, 2011, pp. 32 - 41, ISSN Print: 0976-6545, ISSN Online: 0976-6553.
[20] Devendra Mittal, Om Prakash Mahela and Rohit Jain, “Detection and Analysis of Power
Quality Disturbances Under Faulty Conditions in Electrical Power System”, International
Journal of Electrical Engineering Technology (IJEET), Volume 4, Issue 2, 2013,
pp. 25 - 36, ISSN Print: 0976-6545, ISSN Online: 0976-6553.