Comput. Sci. Appl. Volume 1, Number 3, 2014, pp. 152-157 Received: July 10, 2014; Published: September 25, 2014 Computer Science and Applications www.ethanpublishing.com A Fast and Accurate Fault Detection Approach in Power Transmission Lines by Modular Neural Network and Discrete Wavelet Transform Ömer Faruk Ertuğrul 1 and Muhammet Bahattin Kurt 2 1. Department of Electrical and Electronics Engineering, Batman University, Batman 72060, Turkey 2. Department of Electrical and Electronics Engineering, Dicle University, Diyarbakır 21100, Turkey Corresponding author: Ömer Faruk Ertuğrul (omerfaruk.ertugrul@batman.edu.tr) Abstract: Isolating a faulted line from high voltage power transmission system as quick as possible is important for system security. The aim of this study is to propose a fast and accurate fault detection approach. In this method, firstly the electrical signals were filtered by a designed filter; secondly the wavelet energy of them is computed. Finally the extracted features are analyzed by the modular PNN (probabilistic neural network). The modular PNN was consisted of four different PNN modules, a sorting unit and a fault statistic unit. The mean fault detection duration is 1.65 ms, the accuracy of detecting faults is 98.68% and they are acceptable with respect to the previous works. Additionally, in this study real electrical power transmission line signals, which were collected from digital relays, were used. Key words: Power transmission lines, discreet wavelet, modular, probabilistic neural network. 1. Introduction Electricity is delivered to cities from substations, which are the edges of power transmission systems. For supplying electricity with minimized interrupts, near all substations are connected with others by more than one power transmission line and they formed interconnected power transmission systems. The protecting systems, relays, must isolate the substation from any faulted line as quick as possible to preserve the stability of the rest of the system, preventing system from more complex faults such as voltage sag, switching off large loads or voltage swell and to maintain the system integrity, which is important for a continuous electric power supply. The faults occur when two or more conductors contacted with ground or each other. Fault detection and isolating the faulted line quickly is important because faulted lines may cause accidents which may harm human, damage the equipments or decrease the economical life of devices because of short circuit current. Additionally, the faults may lead to new faults in other power transmission lines or substations. Furthermore, the interrupts will decrease the quality electricity supplied. Therefore, many approaches were introduced to solve this issue. These approaches can be categorized into (a) circuit based methods that detects faults through the voltage, current and impedance changes [1], (b) travelling wave based methods that analyses faults by using the return time of the pulse wave [2], (c) statistical methods that determines faults by using the relative statistical features of voltage and currents and their previous values [3] and (c) expert systems [4-9], such that ANN (artificial neural network), FL (fuzzy logic), SVM (support vector machines). For the last twenty years the expert systems become popular depending on
A Fast and Accurate Fault Detection Approach in Power Transmission Lines 153 being easily adapted to any transmission line, its high accuracy, and generalization ability. The aim of this paper is to evaluate and validate a new approach to detect faults by using wavelet energy and modular probabilistic neural network. This approach is based on the fact that the faults in a transmission line, in general, are the same type which may cause because of workmanship errors, geological facts, strikes, birds, wind, ice load or deformation of insulator materials. The results were showing that the proposed approach is fast and has high accuracy in the fault detection. The obtained results were compared with the results of previous works and fault records of digital relays. The rest of the paper is organized as follows: Section 2 discusses the dataset employed in this study; Section 3 introduces the employed methods; wavelet energy and modular probabilistic neural network; Section 4 describes the proposed approach; Section 5 presents the results and discussions; Section 6 concludes the paper. faulty signals (48 phase A fault, 48 phase B fault, 18 phase C fault, 90 ground fault) and the other 616 is usual electrical signals. An A phase faulted electrical signal sample is shown in Figs. 1 and 2. 3. Employed Methods Wavelet Energy is calculated by adding the square of detailed wavelet coefficients and calculated as [11, 12] ( ) = (1) where denotes scaling coefficients and ( ) is the signal and is its detailed wavelet coefficient. Using small artificial neural networks and combine them to form modular artificial neural networks to solve the problem, is faster than using complex, having too many layer artificial neural networks. 2. Material Generally, the dataset, which is simulated by EMTP, ATP, PSCAD or MATLAB, is used to analyze new approaches [2, 5, 6, 8, 10], therefore, higher accuracy may be obtained. In this study, real electrical power transmission signals, which were recorded by distance protection relays, was used. This relay analyses the voltage and current signals of its integrated transmission line and when it detects a disturbance, it sends a trip signal to a circuit breaker in order to disconnect the faulted line for the rest of the power transmission system can continue to work properly. The digital relays also have the ability of recording the faulty signals in a disturbance case. The power transmission line fault dataset consists of 820 electrical signal set, which is formed with 4 current (IA, IB, IC and IG) and 4 voltage (VA, VB, VC and VG) signals and shows 20 ms time interval with 19,200 Hz sampling frequency. 204 set of it is Fig. 1 Fig. 2 The currents of faulted power transmission line. The voltages of faulted power transmission line.
154 A Fast and Accurate Fault Detection Approach in Power Transmission Lines Modular artificial neural networks are used to simplify complex problems by dividing it into small problems, so getting results more quickly and effectively. In this study, probabilistic neural network [8, 13, 14], which has a high training speed compared to back propagation networks and robust to noise, was used in modules. PNN (probabilistic neural network) is depending on the probability density estimation and it produces results with the concept of winner takes all attitudes. It combines the Bayes decision-making strategy with a nonparametric estimator to obtain the probability density function. It consists four layers, which are input, pattern, summation, and output layers. The output of the pattern layer is calculated by [13] = ( ) ( ) (2) where shows the smoothing parameter, indicates the neuron vector and denotes the dimension of pattern vector. The summation layer neurons compute the maximum likelihood of pattern by ( ) = ( ) ( ) (3) where is tptal number of samples. Finally, the classification is done in decision layer based on Bayes rule as ( ) = ( ) (4) where is the total number of classes. 4. Proposed Approach and frequency response of designed filter are illustrated in Fig. 4. The filter strength the 50 Hz frequency component, because in our country the electricity is transmitted at 50 Hz. Additionally, it must pass 35-65 Hz components while filtering electricity harmonics because in our experiments we observed that the frequencies of faulted electrical signals are varied ± 30%. An A phase faulted signal and filtered signal is shown in Figs. 5 and 6, respectively. DWE (discreet wavelet energy) is used for analyzing transient current and voltage signals while reducing data size (reduces data size by 88% from 256 1 to 32 1 vector size) to have more accurate and faster classification. The DWE was used as a feature extraction method in this study, since the faulty signals have high frequency components; its energy is a distinctive feature. In this study, the fifth level Daubechies-5 discrete wavelet energy is of signals were computed. An A phase current signal and its wavelet energy are shown in Fig. 7. MPNN (modular probabilistic neural networks), is used for having faster training stage with higher accuracy. The algorithm used in this paper is shown in Fig. 8. As seen in Fig. 8, 4 PNN, which formed the MPNN, were used for fault detection. MA, MB, MC and MG modules are created for detecting faults of a phase A, B, C and ground, respectively. Also the faults detected The structure of the overall approach is shown in Fig. 3. Digital filter is used to filter noise, unwanted components such as DC components and also for strength the relevant frequency components. The time Preprocess Feature Extraction Classification Preprocess: Filtering signals Feature Extraction: Wavelet Energy Classification: Modular Probabilistic Neural Network Fig. 3 The procedure of employing the proposed approach. Fig. 4 The filter: (a) time and (b) frequency domain.
A Fast and Accurate Fault Detection Approach in Power Transmission Lines 155 Fig. 5 (a) A signal sample and (b) filtered sample. Fig. 6 The output of filter: (a) in time, (b) in frequency domain. Fig. 8 The algorithm of proposed approach. controlled the order of modules to be employed. 5. Results and Discussion Fig. 7 The filtered signal: (a) in time domain and (b) its wavelet energy. by MA, MB, MC and MG modules are analyzed by a statistical function. Statistical function sorts the phases depend on the occurred fault rates. The Sorting Unit is The obtained accuracy rate was 98.68% in detecting faults by using 10 folds cross validation. Modular artificial neural network increases the success rate of detecting fault, instead of using an ANN, four different PNN modules were used. The extracted features were also classified with popular machine
156 A Fast and Accurate Fault Detection Approach in Power Transmission Lines learning methods and the obtained accuracies were sorted in Table 1. The obtained accuracies in previous studies were between 45%-100% [1 3, 6-11, 14]; it is 98.68% in this study, which is considered in an acceptable range. Dalstein and Kulicke reported that the relays must detect the faults in 10 ms, in very fast relaying systems [16] and the mean fault detection duration is 1.65 ms in proposed method which is in this range. Also the fault detection speed of this approach was compared with the results of previous studies and they are sorted in Table 3. In order to increase the speed of fault detection just one cycle (20 ms) of fault data is used for analyzing. Some distance protection relay used in power transmission system, needs 40 ms for fault detecting. There are two fault records of two different distance protection relay in the same substation, and shown in Fig. 9. In Fig. 9, there are two fault records which are ABG fault (recorded at 10:24, Dec. 23, 2007) and ABCG fault (recorded at 23:21, Jun. 12, 2007). As seen in figures the faults detected in around to 40 ms after the fault occurred. 6. Conclusions This paper presents a fast and accurate method for detecting faults by using computers with expert systems instead of distance protection relays. For having a faster fault detection one cycle of real electrical signal are filtered, and the discrete wavelet energy of filtered signals was used by the modular probabilistic neural network for fault detecting. The obtained accuracy is 98.68%, which is in an acceptable range, with respect to the accuracy results of popular machine learning methods and previous studies, since real electrical signals were used in this study. Additionally, the fault detection speed of the proposed approach is high compared with previous works and faults records of digital relay. Furthermore, only one cycle electrical signals were used in this analysis. Table 1 The obtained classification accuracies (%). Machine learning method Accuracy (%) knn 91.66 Feed forward ANN 88.78 MPNN 98.68 SVM 91.22 Fuzzy logic 90.23 Decision tree (PART) 89.39 Table 2 The signal durations for fault detection. Fault type Detect time (ms) Employed modules One module 1.1 1 AG 1.1 1 BG 1.1 1 CG 1.1 1 ABG 1.1 1 ACG 1.1 1 BCG 1.1 1 AC 2.2 2 AB 2.2 2 ABC 2.2 2 BC 3.3 3 No fault 4.4 4 Mean 1.65 - Table 3 The fault speeds comparison. Time (ms) Used data Sampling frequency Data type Used method In Ref. [15] 1,624 (mean) - 1920 Hz ATP Circuit based In Ref. [16] 5-7 (mean) 100 cycle - ATP ANN In Ref. [17] 4-5 (mean) - 1000 Hz NETOMAC ANN In Ref. [18] 10 (max) 20 cycle 4500 Hz EMTP Fuzzy wavelet In Ref. [19] 10 (max) - EMTPD, PSCAD Fuzzy ANN In this study 1,65 (1,1-3,3) 1 cycle 19200 Real Data MPNN wavelet
A Fast and Accurate Fault Detection Approach in Power Transmission Lines 157 (a) (b) Fig. 9 The signals of faulted power transmission line, recorded by distance protection relays. References [1] X. Wei, J. Liu, Y. Li, Simulation and analysis of power transmission line model, in: Proceedings of 2006 IEEE International Symposium on Power Line Communications and Its Applications, 2006, pp. 256-260. [2] J. Duan, B.H. Zhang, H. Ha, A novel approach to faulted-phase selection using current traveling waves and wavelet analysis, in: Proceedings of IEEE International Conference on Power System Technology (PowerCon 2002), 2002, pp. 1146-1150. [3] A.R. Sedighi, M. Haghifam, O.P. Malik, M.H. Ghassemian, High impedance fault detection based on wavelet transform and statistical pattern recognition, IEEE Transactions on Power Delivery 20 (4) (2005) 2414-2421. [4] V. Malathi, N.S. Marimuthu, S. Baskar, Intelligent approaches using support vector machine and extreme learning machine for transmission line protection, Neurocomputing 73 (10) (2010) 2160-2167. [5] K.H. Kashyap, U.J. Shenoy, Classification of power system faults using wavelet transforms and probabilistic neural networks, in: Proceedings of the IEEE 2003 International Symposium on Circuits and Systems (ISCAS'03), 2003, pp. III-423-III-426. [6] V. Malathi, N.S. Marimuthu, Multi-class support vector machine approach for fault classification in power transmission line, in: Proceedings of IEEE International Conference on Sustainable Energy Technologies (ICSET 2008), 2008, pp. 67-71. [7] A. de Souza Gomes, M.A. Costa, T.G.A. de Faria, W.M. Caminhas, Detection and classification of faults in power transmission lines using functional analysis and computational intelligence, IEEE Transactions on Power Delivery 28 (3) (2013) 1402-1413. [8] F. Mo, W. Kinsner, Probabilistic neural networks for power line fault classification, in: Proceedings of IEEE Canadian Conference on Electrical and Computer Engineering, 1998, Vol. 2, pp. 585-588. [9] K.V. Babu, M. Tripathy, A.K. Singh, Recent techniques used in transmission line protection: A review, International Journal of Engineering, Science and Technology 3 (3) (2011) 1-8. [10] P. Sridhar, V.P.C. Rao, B.P. Singh, Detection of fault in a 400kV transmission line using EMTP and ANN method, International Journal on Engineering Applications (IREA) 2 (1) (2014) 20-24. [11] X. Xu, M. Kezunovic, Automated feature extraction from power system transients using wavelet transform, in: Proc. Int. Conf. Power System Technology, 2002, Vol. 4, pp. 1994-1998. [12] A.M. Gaouda, S.H. Kanoun, M.M.A. Salama, A.Y. Chikhani, Pattern recognition applications for power system disturbance classification, IEEE Power Engineering Review 22 (1) (2002) 69-70. [13] S. Deepa, V.S. Bharathi, Textural feature extraction and classification of mammogram images using CCCM and PNN, IOSR Journal of Computer Engineering (IOSR-JCE) 10 (6) (2013) 7-13. [14] P. Chiradeja, A. Ngaopitakkul, Prediction of fault location in overhead transmission line and underground distribution cable using probabilistic neural network, International Review of Electrical Engineering (IREE) 8 (2) (2013) 762-768. [15] J.A. Jiang, C.S. Chen, C.W. Liu, A new protection scheme for fault detection, direction discrimination, classification, and location in transmission lines, IEEE Transactions on Power Delivery 18 (1) (2003) 34-42. [16] T. Dalstein, B. Kulicke, Neural network approach to fault classification for high speed protective relaying, IEEE Transactions on Power Delivery, 10(2) (1995) 1002-1011. [17] M. Oleskovicz, D.V. Coury, R.K. Aggarwal, A complete scheme for fault detection, classification and location in transmission lines using neural networks, in: Proceedings of 7th International Conference on Developments in Power System Protection, 2001,pp. 335-338. [18] O.A. Youssef, Combined fuzzy-logic wavelet-based fault classification technique for power system relaying, IEEE Transactions on Power Delivery 19 (2) (2004) 582-589. [19] H. Wang, W.W.L. Keerthipala, Fuzzy-neuro approach to fault classification for transmission line protection, IEEE Transactions on Power Delivery 13 (4) (1998) 1093-1104.