combination of discrete wavelet transform and ... - Semantic Scholar

International Journal of Innovative Computing, Information and Control Volume 7, Number 4, April 2011

c ICIC International ⃝2011 ISSN 1349-4198 pp. 1861–1873

COMBINATION OF DISCRETE WAVELET TRANSFORM AND PROBABILISTIC NEURAL NETWORK ALGORITHM FOR DETECTING FAULT LOCATION ON TRANSMISSION SYSTEM Atthapol Ngaopitakkul and Chaiyan Jettanasen Faculty of Engineering King Mongkut’s Institute of Technology Ladkrabang Chalongkrung Rd., Ladkrabang, Bangkok 10520, Thailand [email protected]

Received December 2009; revised May 2010 Abstract. This paper proposes a new algorithm for detecting faults in an electrical power transmission system, using discrete wavelet transform (DWT) and probabilistic neural network (PNN). Fault conditions are simulated using ATP/EMTP to obtain current signals. The algorithm used to analyze fault locations is developed on MATLAB. Fault detection is processed using the positive sequence current signals. The comparison among the maximum coefficients in first scale of each bus, which can detect fault, is performed in order to detect the faulty bus. The first peak time obtained from the faulty bus is used as an input for training pattern. Various cases based on Thailand electricity transmission systems are studied to verify the validity of the proposed technique. The result shows that the algorithm is capable of performing the fault locations with accuracy. Keywords: Discrete wavelet transform, Fault location, Probabilistic neural network

1. Introduction. Nowadays, transmission lines are more complicated as a large grid owing to increasing demand of electric power. In an interconnected-electrical transmission system, a precise protection scheme is required in order to ensure the extreme level of the system reliability. Generally, when fault occurs on transmission lines, detecting fault is very necessary in order to clear fault before it generates the damage to the power system. The traditional method of signal analysis is based on Fourier transform, but the fault signals are non-stationary transient so the signal analysis methods with Fourier transform are not quite efficient. Recently, the development of an algorithm for detecting faults in the transmission lines has been progressed, resulting in transient-based techniques [1]. For the transient-based protection to be accurately applied in operation, the application of wavelet transform is used [1-9]. The advantage of the wavelet transform is that the band of analysis can be adjusted to allow high-frequency and low-frequency components to be precisely detected. As a result, the wavelet transform is not intended to replace the Fourier transform in analyzing steady state signals. It is an alternative tool for analyzing non-stationary or non-steady state signals. This is due to that the wavelet transform is very effective in detecting transient signals generated by the faults. The wavelet transform was initially proposed by Magnago et al. [10]. In the literature for fault location, most researches [4,10-16] have only considered the fault location for single bus and two-bus systems but not for multi-terminal. The location of the fault was normally calculated using travelling wave approach, as presented in [10]. In addition, artificial intelligence (AI) has been also reported in the literature for fault location. In [11], the paper describes an artificial neural network-based algorithm for fault location. The inputs are phasors of pre-fault and superimposed voltages and currents from all phases 1861

1862

A. NGAOPITAKKUL AND C. JETTANASEN

of the transmission line. Nowadays, fault diagnosis for the transmission line has also been progressed with the applications of wavelet transform (WT) and artificial intelligent [17-19]. In [17], the paper proposes a fault location method employing wavelet fuzzy neural network to use post-fault transient and steady-state measurements. Several decision algorithms for locating fault on multi-terminal have been proposed [20,21], but their solutions and techniques are different. In [21], this paper presents a new method for locating faults on three-terminal power lines using three-phase current from all three-terminals and additionally three-phase voltage from the terminal at which a fault locator is installed. Conventional method for fault location, employed by Electricity Generating Authority of Thailand (EGAT), is the Line Fault Locator (LFL) Type “c”. As the devices of LFL are complicated and expensive, a new technique has to be investigated. Therefore, this paper is focused on the ANNs algorithm for the location of fault along the transmission systems in order to identify the fault location. This is due to that ANNs are a useful tool for solving and selecting a precision algorithm for a protection unit. There are currently many types of ANNs being commercially used. Back-propagation neural network (BPNN) is a type of ANNs, which is widely applied in such a system today. However, it is partly limited by the slow training performance. This drawback should be improved; otherwise the other types of neural network would be developed instead. Finally, the probabilistic neural network (PNN) is selected in the algorithm because it uses less training data and time compared with BPNN. Although, the PNN has not been yet fully evaluated in comparison to BPNN however the PNN approach offers several major advantages such as rapid training, added or deleted data from training set without lengthy retaining, and etc. As a result, it is useful to be able to perform fault location on the transmission line using wavelet transform and PNN. This paper is aimed to propose a combination of wavelet transforms and PNN to detect the faults on transmission systems. The fault conditions will be simulated using ATP/EMTP. The analysis and diagnosis were performed using MATLAB on a PC Pentium IV 2.4 GHz 512 MB. The systems under consideration have a radial and loop structure in order to show the advantage of the proposed method. Fault signals in each case are extracted to several scales on the wavelet transforms, and then are used as an input for a training process on the neural networks. A new technique to identify fault locations on the transmission system is discussed. In addition, the construction of the decision algorithm is detailed and implemented with various case studies, based on Thailand electricity transmission systems. 2. Power System Simulation Using EMTP. The ATP/EMTP [22] is employed to simulate fault signals at a sampling rate of 200 kHz (corresponding to the chosen sampling time used in ATP/EMTP, which is equal to 5 µs). The fault types are chosen based on the Thailand’s transmission system as shown in Figures 1 and 2. To avoid complexity, the fault resistance is assumed to be 10 Ω. Fault patterns in the simulation are performed with various changes of system parameters as follows: - Fault types under consideration, namely: single phase to ground (SLG: AG, BG, CG), double-line to ground (DLG: ABG, BCG, CAG), line to line (L-L: AB, BC, CA) and three-phase fault (3-P: ABC). - Fault locations on the each transmission lines were at the distance of 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80% and 90%, measured from the sending end. - Inception angle on a voltage waveform was varied between 0◦ – 330◦ , with the increasing step of 30◦ . Phase A was used as a reference. The example of original and ATP/EMTP simulated fault signals for phase A to ground fault (AG) in each phase at the sending end (MM3) of the transmission lines is illustrated

WAVELET TRANSFORM AND PROBABILISTIC NEURAL NETWORK ALGORITHM

1863

Figure 1. The system used in simulations studies for single circuit structure (System 1) [23]

Figure 2. The system used in simulations studies for loop structure (System 2) [23] in Figure 3(a) and Figure 3(b) respectively. This is a fault occurring in phase A to ground (AG) at the length of 35% measured from the bus MM3 as depicted in Figure 1. The similarity between the original and simulated fault signals waveforms can be seen obviously. The fault signals generated using ATP/EMTP are interfaced to the MATLAB in order to analyse the transient high frequency components by using wavelet toolbox. 3. Fault Detection Algorithm. Fault detection decision algorithm is processed using the positive sequence current signals as illustrated in Figure 4. The fault signals generated using ATP/EMTP, is extracted to several scales with the Wavelet transform. The mother wavelet, daubechies4 (db4) [4,19,24], is employed to decompose high frequency components from the signals. Coefficients obtained using DWT of signals, then, are squared so that the abrupt change in the spectra can be clearly found. It is obviously seen that when fault occurs, the coefficients of high frequency components have a sudden change compared with those before an occurrence of the faults as illustrated in Figure 5. After applying the Wavelet transform to the positive sequence currents, the comparison of the

1864

A. NGAOPITAKKUL AND C. JETTANASEN

(a)

(b) Figure 3. (a) Example of original fault signals for AG fault; (b) Example of ATP/EMTP simulated fault signals for AG fault

WAVELET TRANSFORM AND PROBABILISTIC NEURAL NETWORK ALGORITHM

1865

coefficients from each scale is considered. The fault detection can be then analyzed, as illustrated in Figure 4.

Figure 4. Flowchart for fault detection

1866

A. NGAOPITAKKUL AND C. JETTANASEN

Figure 5. Wavelet transform from scale 1 to 5 for the positive sequence of current signals shown in Figure 3

Figure 6. Example of wavelet transform for the positive sequence of phase A to ground fault at the transmission system (section WN-CBG) where, WN1T, WN2T are WN bus section WN-CBG circuit 1 and circuit 2 respectively.

WAVELET TRANSFORM AND PROBABILISTIC NEURAL NETWORK ALGORITHM

1867

WN1O, WN2O are WN bus section WN-SNO circuit 1 and circuit 2 respectively. CBG1T, CBG2T are CBG bus section WN-CBG circuit 1 and circuit 2 respectively. CBG1N, CBG2N are CBG bus section SNO-CBG circuit 1 and circuit 2 respectively. SNO1O, SNO2O are SNO bus section WN-SNO circuit 1 and circuit 2 respectively. SNO1N, SNO2N are SNO bus section SNO-CBG circuit 1 and circuit 2 respectively. From Figure 5 and Figure 6 show that the coefficient at each scale of the wavelet transform does change. Therefore, the result obtained from the fault detection algorithm presumes that these signals are in their fault condition. However, when carefully considering, Figure 6 is found that all coefficients obtained from the positive sequence currents at every bus have a change of more than 5 times of a normal value during the faults due to the effect of a loop structure of the transmission network. The comparison among the maximum coefficients in first scale of each bus [25], which can detect fault, is carried out in order to detect the faulty bus. In case of double circuit, the maximum coefficients obtained from same buses are also compared in order to detect the faulty circuit. The first peak times obtained from the faulty bus are used as an input data for neural network as shown in Figures 7 and 8.

Figure 7. First peak in the scale 1 at both ends of transmission lines for the positive sequence of current signal shown in Figure 4 4. Neural Network Decision Algorithm and Results. Probabilistic neural network (PNN) is developed by Donald Specht, to perform pattern classification using Gaussian potential functions and Bayes decision theory [26]. The PNN consists of three layers which are an input layer, a hidden radial basis layer and a competitive layer as illustrated in Figure 9. Each layer is interconnected by weights. Radial basis function and competitive function, which are activation function, are comprised in hidden radial basis layer and competitive layer respectively. Moreover, the number of neurons in radial basis layer is always equal to the number of training sets.

1868

A. NGAOPITAKKUL AND C. JETTANASEN

Figure 8. The first peak in the scale 1 at the faulty bus for the signal shown in Figure 5

Figure 9. Probabilistic neural network [27] A training process of PNN [27] involves two stages as follows: 1. Input values are propagated to each neuron in the first layer. The radial basis layer computes distance from the input vector to weight vector, and produces output in radial basis layer as in Equation (1). ) ( ∥p − IW1,1 ∥2 (1) φ (p) = exp − σj2 where, p is the input pattern vector, IW 1,1 is the center vector of radial basis layer, σ is the spread constant for radial basis layer, which corresponds to bias value (b = φ(p) is the output of radial basis layer.

0.8326 ), Spread

WAVELET TRANSFORM AND PROBABILISTIC NEURAL NETWORK ALGORITHM

1869

2. Each neuron in the competitive layer receives all radial basis layer outputs associated with a given class, and produces, as its net output, a vector of probabilities. Finally, a competitive activation function on the output of the competitive layer picks the maximum of these probabilities, and produces a 1 for that class and a 0 for the other classes as shown in Equation (2) [27]. o/pAN N = f 4 (LW2,1 ∗ φ (p)) (2) where, LW 2,1 = weight vector between radial basis layer and competitive layer, f 4 = competitive activation function. From the simulated signals, the coefficients of scale 1, which are obtained using the wavelet transforms, are used for training and test processes of the PNN. Input data sets are normalized and divided into 4176 sets for training, and 2736 sets for tests. Before the training process, PNN structure consists of 2 neurons input and 1 neuron output while the number of neurons in radial basis layer are 4176 neurons (due to that number of neurons is always equal to the number of training sets). The inputs pattern is the first peak time in first scale of faulty buses at 1/4 cycle of positive sequence for post-fault currents as mentioned in the previous section. The output variables of the PNN are designated range 1 to 9, corresponding to various locations of faults as shown in Table 1. Table 1. Output patterns from neural networks Fault location Distance measured Distance measured from the senOutput (Distance measured from the sending ding end for loop structure (km) of PNN from the sending end) end for radial (%) structure (km) Section T Section O Section N 1 10% 32.5 18.3 6.2 10.4 2 20% 65 36.6 12.4 20.8 3 30% 97.5 54.9 18.6 31.2 4 40% 130 73.2 24.8 41.6 5 50% 162.5 91.5 31 52 6 60% 195 109.8 37.2 62.4 7 70% 227.5 128.1 43.4 72.8 8 80% 260 146.4 49.6 83.2 9 90% 292.5 164.7 55.8 93.6 Figure 10 shows an algorithm used in the training process for the PNN. During the training process, PNN begins with the random initial weight and, increasing spread in 0.8326 radial basis layer which corresponds to bias value (b = Spread ) from 0.0001 until 0.1. The step of increase is at 0.0001 to compute the minimum value of MAPE as shown in Equation (3). This procedure is repeated until the maximum number of spread is reached, or MAPE of test set is equal to zero. Results from the training process are shown in Table 2. n 1 ∑ o/pAN N i − o/pT ARGET i (3) M AP E = ∗ ∗ 100% n o/pT ARGET i i=1

where, n = number of test set. After the training process, the algorithm was employed in order to calculate the distance of fault along the transmission systems. Case studies were varied to verify the algorithm capability. The system under consideration is shown in Figures 1 and 2. The fault location is defined at the distance between 10% and 90% of the transmission length measured from

1870

A. NGAOPITAKKUL AND C. JETTANASEN

Figure 10. Process of training

WAVELET TRANSFORM AND PROBABILISTIC NEURAL NETWORK ALGORITHM

1871

Table 2. Results of training process Information for comparison Radial structure Loop structure Spread 0.0442 0.001 Iterations 1000 1000 Number of Error for Test set 0 0 Best time of training process (second) 2.3 4.32 Total time of training process (minute) 3.4 12.09 the sending end and receiving end respectively. Various case studies were performed with various types of faults at each location on the transmission network including the variation of fault inception angles and locations on each transmission line. The total number of the case studies was 8,208. The comparison between an average error in fault locations obtained from the PNN algorithm proposed in this paper and that of the former wavelet algorithm developed by Markming et al. [4] is shown in Table 3. It can be seen that the new algorithm can provide a better performance in predicting the fault locations. Table 3. Comparison of average error for fault locations at various types of faults Average error (km) Types of Number of Fault Wavelet based on faults case studies detection DWT and PNN Traveling wave [4] SLG 324 100% 0.038 0.855 DLG 324 100% 0.017 0.845 MM3-TTK L-L 324 100% 0.030 0.859 ABC 108 100% 0.017 0.785 SLG 648 100% 0.00 0.71 DLG 648 100% 0.00 0.71 WN-CBG L-L 648 100% 0.00 0.71 ABC 432 100% 0.00 0.71 SLG 648 100% 0.00 0.27 DLG 648 100% 0.00 0.27 WN-SNO L-L 648 100% 0.00 0.27 ABC 432 100% 0.00 0.27 SLG 648 100% 0.00 0.41 DLG 648 100% 0.00 0.41 SNO-CBG L-L 648 100% 0.00 0.41 ABC 432 100% 0.00 0.41 Average 8208 100% 0.01 0.56 Section

5. Conclusions. This paper proposes an algorithm based on a combination of DWT and PNN algorithm to identify fault location on the transmission systems. Daubechies4 (db4) is selected as a mother wavelet. The DWT has been employed to decompose high frequency components from fault signals. Positive sequence current signals are used in fault detection. The maximum coefficients of the positive sequence current obtained from all buses are compared in order to detect the faulty bus on the transmission system. It is found that the fault detection algorithm can detect fault with the accuracy of 100% using scale 1 only. PNN has been selected in the decision algorithm for predicting the location of fault. The first peak times obtained from the faulty bus are used as an input for the training process of a neural network in a decision algorithm. The results show

1872

A. NGAOPITAKKUL AND C. JETTANASEN

that the proposed algorithm is able to detect the faulty bus with the accuracy of 100% and identify fault location with the average error of 0 km. Acknowledgment. This work is partially supported by King Mongkut’s Institute of Technology Ladkrabang Research fund. The authors also gratefully acknowledge the helpful comments and suggestions of the reviewers, which have improved the presentation. REFERENCES [1] Z. Q. Bo, F. Jiang, Z. Chen, X. Z. Dong, G. Weller and M. A. Redfern, Transient based protection for power transmission systems, IEEE Power Engineering Society Winter Meeting, vol.3, pp.1832-1837, 2000. [2] C. H. Kim and R. Aggarwal, Wavelet transforms in power systems. I. General introduction to the wavelet transforms, IEE Power Engineering Journal, vol.14, no.2, pp.81-87, 2000. [3] O. A. S. Youssef, Fault classification based on wavelet transforms, IEEE/PES Transmission and Distribution Conference and Exposition, vol.1, pp.531-538, 2001. [4] P. Maknimg, S. Bunjongjit, A. Kunakorn, S. Jiriwibhakorn and M. Kando, Fault diagnosis in transmission lines using wavelet transform analysis, IEEE/PES Transmission and Distribution Conference and Exhibition 2002: Asia Pacific, vol.3, pp.2246-2250, 2002. [5] A. Ngaopitakkul and A. Kunakorn, Internal fault classification in transformer windings using combination of discrete wavelet transforms and back-propagation neural networks, International Journal of Control, Automation, and Systems, vol.4, no.3, pp.365-371, 2006. [6] Z. Zhang, Y. Ohara, H. Toda, T. Miyake and T. Imamura, De-noising method by combining adaptive line enhancer and complex discrete wavelet transform, ICIC Express Letters, vol.1, no.2, pp.145-151, 2007. [7] S. P. Lee and C. H. Loh, Object-oriented design metrics as early quality indicators of faulty classes and components, ICIC Express Letters, vol.3, no.3(A), pp.525-530, 2009. [8] O. Mustapha, D. Lefebvre, M. Khalil, G. Hoblos and H. Chafouk, Fault detection algorithm using DCS method combined with filters bank derived from the wavelet transform, International Journal of Innovative Computing, Information and Control, vol.5, no.5, pp.1313-1328, 2009. [9] I. Usman, A. Khan, A. Ali and T.-S. Choi, Reversible watermarking based on intelligent coefficient selection and integer wavelet transform, International Journal of Innovative Computing, Information and Control, vol.5, no.12(A), pp.4675-4682, 2009. [10] F. H. Magnago and A. Abur, Fault location using wavelets, IEEE Trans. on Power Delivery, vol.13, no.4, pp.1475-1480, 1998. [11] Z. Chen and J.-C. Maun, Artificial neural network approach to single-ended fault locator for transmission lines, IEEE Trans. Power Systems, vol.15, pp.370-375, 2000. [12] A. J. Mazon, I. Zamora, J. F. Minambres, M. A. Zorrozua, J. J. Barandiaran and K. Sagastabeitia, A new approach to fault location in two-terminal transmission lines using artificial neural networks, Electric Power Systems Research, vol.56, pp.261-266, 2000. [13] T. Kawady and J. Stenzel, A practical fault location approach for double circuit transmission lines using single end data, IEEE Trans. on Power Delivery, vol.18, no.4, pp.1166-1173, 2003. [14] J. Izykowski, E. Rosolowski and M. M. Saha, Locating faults in parallel transmission lines under availability of complete measurements at one end, IEE Proc. on Generation, Transmission and Distribution, vol.151, no.2, pp.268-273, 2004. [15] G. Song, J. Suonan, Q. Xu, P. Chen and Y. Ge, Parallel transmission lines fault location algorithm based on differential component net, IEEE Trans. on Power Delivery, vol.20, no.4, pp.2396-2406, 2005. [16] L. S. Martins, J. F. Martins, F. V. Pires and C. M. Alegria, A neural space vector fault location for parallel double-circuit distribution lines, International Journal of Electrical Power & Energy Systems, vol.27, no.3, pp.225-231, 2005. [17] C. Fan, K. K. Li, W. L. Chan, W. Yu and Z. Zhang, Application of wavelet fuzzy neural network in locating single line to ground fault (SLG) in distribution lines, International Journal of Electrical Power & Energy Systems, vol.29, no.6, pp.497-503, 2007. [18] M. J. Reddy and D. K. Mohanta, A wavelet-fuzzy combined approach for classification and location of transmission line faults, Electrical Power and Energy Systems, vol.29, no.9, pp.669-678, 2007.

WAVELET TRANSFORM AND PROBABILISTIC NEURAL NETWORK ALGORITHM

1873

[19] P. Chiradeja and A. Ngaopitakkul, Identification of fault types for single circuit transmission line using discrete wavelet transform and artificial neural networks, Proc. of 2009 International Conference on Electrical Engineering, Hong Kong, 2009. [20] Y.-H. Lin, C.-W. Liu and C.-S. Yu, A new fault locator for three-terminal transmission lines-using two-terminal synchronized voltage and current phasors, IEEE Trans. on Power Delivery, pp.452-459, 2002. [21] J. Izykowski, E. Rosolowski, M. M. Saha, M. Fulczyk and P. Balcerek, A fault-location method for application with current differential relays of three-terminal lines, IEEE Trans. on Power Delivery, vol.22, no.4, pp.2099-2107, 2007. [22] D. V. Dommelen, Alternative Transient Program Rule Book, Leuven EMTP Center, Belgium, 1991. [23] Switching and Transmission Line Diagram, Electricity Generation Authorisation Thailand (EGAT). [24] N. S. D. Brito, B. A. Souza and F. A. C. Pires, Daubechies wavelets in quality of electrical power, Proc. of IEEE International Conference on Harmonics and Quality of Power, pp.511-515, 1998. [25] T. Patcharoen, A. Ngaopitakkul and A. Kunakorn, Identification of fault types for a three-bus transmission network using discrete wavelet transform and probabilistic neural networks, Proc. of the 8th International Power Engineering Conference, Singapore, pp.1137-1142, 2007. [26] S. Haykin, Neural Networks: A Comprehensive Foundation, Prentice-Hall, 1999. [27] H. Demuth and M. Beale, Neural Network Toolbox User’s Guide, The Math Work Inc., 2001.