B. Gabrys, R.J. Howlett, and L.C. Jain (Eds.): KES 2006, Part I, LNAI 4251, pp. 179–186, 2006. © Springer-Verlag Berlin Heidelberg 2006
Power Transformer Fault Diagnosis Using
Support Vector Machines and Artificial Neural
Networks with Clonal Selection Algorithms Optimization
Ming-Yuan Cho, Tsair-Fwu Lee, Shih-Wei Gau, and Ching-Nan Shih
Department of Electrical Engineering, National Kaohsiung University of Applied Science, Kaohsiung, Taiwan 807, ROC
mycho@mail.ee.kuas.edu.tw, tflee@bit.kuas.edu.tw, stone@cc.kuas.edu.tw
Abstract. This paper presents an innovative method based on Artificial Neural
Network (ANN) and multi-layer Support Vector Machine (SVM) for the pur-pose of fault diagnosis of power transformers. A clonal selection algorithm (CSA) based encoding technique is applied to improve the accuracy of classifi-cation, which demonstrated in the literature for the first time. With features and RBF kernel parameters selection to predict incipient fault of power transformer improve the accuracy of classification systems and the generalization perform-ance. The proposed approach is distinguished by removing redundant input fea-tures that may be confusing the classifier and optimizing the selection of kernel parameters. Simulation results of practice data demonstrate the effectiveness and high efficiency of the proposed approach, which makes operation faster and also increases the accuracy of the classification.
1 Introduction
This Paper proposes a combination method to improve support vector machines (SVM) based fault diagnosis for power transformers through the operation of clonal selection algorithm (CSA). The SVM is a relatively new kind of learning machine and soft computing method based on the structural risk minimization (SRM) principle and statistical machine learning theory (SLT) that performs structural risk minimization on a nested set structure of separating hyperplanes which was proposed by Vapnik and his group at AT&T Bell Laboratories [1]. SVMs have been introduced as a new methodology for solving classification and functional regression with many success-ful applications [2]. It is powersuccess-ful for the problem with small sampling, nonlinear and high dimension. It provides a unique solution and is a strongly regularized method appropriate for ill-defined problems.
Power transformers are definitely vital devices in a transmission and distribution system, as expensive items; need to be carefully monitored throughout their operation. Fault diagnosis of it is important for safety of the device and relevant power system. To predict incipient fault is gaining importance in industry because of the failure of a power transformer may cause an interruption in power supply and loss of revenue. Therefore, it is of great importance to detect incipient failures in power transformers
180 M.-Y. Cho et al.
as early as possible, so that we can switch them safely and improve the reliability of power systems [3]. The most widely applied method for incipient faults in power transformers can be detected and monitored based on dissolved gas analysis (DGA) which has been proved that are related closely to transformer’s internal faults [3]. Fault gases are produced by degradation of the transformer oil and solid insulating materials, such as paper, pressboard and transformer board, which are all made of cellulose. According to the IEC 599 [4], the fault related gases include hydrogen (H2),
oxygen (O2), nitrogen (N2), methane (CH4), ethylene (C2H4), acetylene (C2H2), eth-ane (C2H6), carbon dioxide (CO2), and carbon monoxide (CO) that can be determined from a DGA test. As study results report, corona or partial discharge, thermal heating and arcing are the three main causes for insulation degradation in a transformer [5]. The energy dissipation is least in corona or partial discharge, medium in thermal heat-ing, and highest in arcing [6].
Various fault diagnosis techniques have been proposed in the literature, including the conventional key gas method, ratio-based method [7] and recently artificial intelligence methods, such as expert system [8], fuzzy logic [9] and artificial neural networks (ANNs) [10], and the hybrid combinations of these methods have given promising results [11]. The conventional key gas and ratio methods are based on experience in fault diagnosis using DGA data, which may vary from utility to utility due to the heuristic nature of the methods and the fact that no general mathematical formulation can be utilized. The expert system and fuzzy logic approaches can take DGA standards and other human expertise to form decision-making system. Infor-mation, such as influence of transformer size, manufacturer, volume of oil, gassing rates and history of the diagnosis result can be utilized. However, there are some intrinsic shortcomings, for example, the difficulty of acquiring knowledge and maintaining a database. Both methods need a large knowledge base that must be constructed manually. In order to overcome the shortcomings, one solution is real-ized to obtain its optimization result through evolutionary algorithm [12]. And a novel evolution computation enhanced method was proposed to ameliorate the fuzzy diagnosis capabilities [13]. The traditional ANN method can directly acquire experience from the training data, and overcome some of the shortcomings of the expert system. However, it suffers from a number of weaknesses, including the need for a large number of controlling parameters, difficulty in obtaining a stable solution and the danger of over-fitting. To overcome the drawbacks of traditional neural networks, a new type of neutral network is proposed for incipient fault diag-nosis of power based on extension theory [7]. As ANNs, expert system and fuzzy logic approaches have their advantages and disadvantages, hybrid artificial intelli-gence approaches are also under consideration. Their disadvantages could be over-come by connecting expert system, fuzzy logic, ANNs and evolutionary algorithms as a whole [14].
Recently, SVM, a novel network algorithm, originally developed by Vapnik and Cortes [15] has emerged as one powerful tool for data analysis. It is powerful for the problem with small sampling, nonlinear and high dimension. SVM has been widely used for many applications, such as face recognition [11], time series forecasting [16], fault detection [17] and modeling of nonlinear dynamic systems [18].
Power Transformer Fault Diagnosis Using SVMs and ANNs 185
Table 4. Performance with CSA optimization for the Taipower Company dataset
With CSA Optimization
Classifier No. of inputs Training success(%) Test success(%) Training time(s) ANN 4 100 99.98 5.335 SVM 4 100 100 0.676
4 Conclusions
In this paper, the selection of input features and the appropriate classifier parameters have been optimized using a CSA-based approach. A procedure is presented for de-tection of power transformer faulty situation using two classifiers, namely, ANNs and SVMs with CSA-based features and parameters selection from DGA inherently im-precise signals. The appropriate radius-margin bound is selected as the objective func-tion for automatic parameters selecfunc-tion for SVM with experiment test. Experiment results of practice data demonstrate the effectiveness and high efficiency of the pro-posed approach. The classification accuracy of SVMs was better than of ANNs, with-out CSA in this application. With CSA-based selection, the performances of both classifiers were comparable at nearly 100%. The results show the potential applica-tion of CSAs for off-line features and parameters selecapplica-tion. This also opens up the potential use of optimized features and classifier parameters for real-time implemen-tation leading to possible development of an automated DGA condition monitoring and diagnostic system. We conclude that our method is not only able to figure out optimal parameters but also minimize the number of features that SVM classifier should process and consequently maximize the diagnosis rate of DGA.
References
1. Vapnik V, Golowich S, Smola A. “Support Vector Method for Function Approximation, Regression Estimation, And Signal Processing [A].” In: Mozer M, Jordan M, Petsche T (eds). Neural Information Processing Systems [M]. MIT Press, 1997, 9.
2. O. Chapelle, V. Vapnik, O. Bousqet, and S. Mukherjee. Choosing Multiple Parameters for Support Vector Machines. Machine Learning, 46(1):131 – 159, 2002.
3. IEC Publication 599, “Interpretation of The Analysis of Gases in Transformers And Other Oil-Filled Electrical Equipment In Service,” Fisrt Edition, 1978.
4. “IEEE Guide for The Interpretation of Gases Generated In Oil-Immersed Transformers,” IEEE Std.C57.104-1991.
5. R.R. Rogers, “IEEE And IEC Codes to Interpret Faults in Transformers Using Gas in Oil Analysis,” IEEE Trans Electron. Insulat. 13 (5), pp.349–354,1978.
6. M.H. Wang, “A Novel Extension Method for Transformer Fault Diagnosis,” IEEE Trans. Power Deliv. 18 (1),pp.164–169,2003.
7. N.C.Wang,”Development of Monitoring and On-line Diagnosing System for Large Power Transformers,”（Power Reserch Institute,TPC）.
8. Yuanping.Ni, “One Intelligent Method for Fault Diagnosis and Its Application,” IEEE Trans. Power Deliv, pp755-758,2000.
186 M.-Y. Cho et al.
9. Ganyun Lv., Haozhong Cheng, Haibao Zhai, Lixin Dong, “Electric Power Systems Re-search 75,” 9–15, 2005 Elsevier B.V. All rights reserved.
10. L.B. Jack, A.K. Nandi, “Fault Detection Using Support Vector Machines and Artificial Neural Networks: Augmented by Genetic Algorithms,”Mech. Syst. Signal Process. 16 (2–3), pp.373–390, 2002.
11. W.C. Chan, C.W. Chan, K.C. Cheung, C.J. Harris, “On The Modeling of Nonlinear Dy-namic Systems Using Support Vector Neural Networks,” Eng. Appl. Artif. Intell. 14, pp.105–113,2001.
12. H. Frohlich. Feature Selection for Support Vector Machines by Means of Genetic Algo-rithms. Master’s thesis, University of Marburg, 2002.
13. M,Dong, ”Fault Diagnosis Model Power Transformer Based on Statistical Learning The-ory and Dissolved Gas Analysis,” IEEE 2004 Internation Symposium on Electrical Insula-tion,p85-88.
14. Y.C. Huang, H.T. Yang, C.L. Huang, Developing a new transformer fault diagnosis sys-tem through evolutionary fuzzy logic, IEEE Trans. Power Deliv. 12 (2) (1997) 761–767. 15. C. Cortes, V. Vapnik, Support-vector Networks, Machine Learn. 20 (3) (1995) 273–295. 16. E.H. Tay Francis, L.J. Cao, Application of support vector machines in financial time series
forecasting, Omega. 29 (4) (2001) 309– 317.
17. ”BECTA”,Power Station Magazine in Russia,No.6,1998.
18. W.W. Yan, H.H. Shao, “Application of Support Vector Machine Nonlinear Classifier to Fault Diagnoses,” Proceedings of the Fourth World Congress Intelligent Control and Auto-mation, 10–14 Shanghai, China, pp. 2697–2670, June 2002.
19. J. H. Lee, and C. J. Lin, Automatic model selection for support vector machines, Technical Report, Dept. of Computer Science and Information Engineering, Taipei, Taiwan, Novem-ber 2000.
20. V.N. Vapnik, “The Nature of Statistical Learning Theory,” Springer-Verlag, New York, 1995.
21. Michel Duval, ”Interpretation of Gas-In-Oil Analysis Using New IEC Publication 60599 and IEC TC 10 Databases,” IEEE Electrical Insulation Magazine, 2001 — Vol. 17, No. 2
