Fault Section Locating In Power Distribution Network Using LS-SVM And IMF Energy Moment

In order to deal with the global energy crisis and environmental problems, the smart power grid concept has come into being as the times requires. Thereinto, intelligentized electric fault protection is an important research direction. Due to low fault currents in power distribution network, current signals are susceptible to interference of external environment. Therefore fault information is hard to be detected. Furthermore, while the fault is present, the potential of other two phases relative to the ground reaches ?3 of the normal operating voltage, creating additional stress for the insulation. Insulation failures may inflict additional ground faults in the system and more likely to develop a multi-phase, multi-point short trouble, enlarge the scope of accident. Thus it is necessary to find a new method to recognize faults efficiently and locate the fault section quickly in complex environments.Firstly, the thesis discusses single phase to ground fault location theory from active and passive form. The advantages and disadvantages of existing fault location theory are explained. In particular, the empirical mode decomposition(EMD) and the Support Vector Machine(SVM) are studied on power system applications. The Ensemble Empirical Mode Decomposition(EEMD) algorithm is proposed for signal processing, which not only preserves the merits of the traditional signal processing method, but also solves mode mixing phenomenon produced by some noise signal. By comparing EEMD to Fourier transform, wavelet transform and the EMD algorithm respectively, the characteristics of different signal extraction algorithm is described in detail. A brief introduction to least squares support vector machine(LS-SVM) is provided and comparing it with BP neural network and support vector machine on classification of fast and accurate aspect. A new method of fault selection based on the intrinsic mode function(IMF) energy moment and LS-SVM is proposed for power system. The localization characteristics of EEMD are used to quantify the fault, and then the LS-SVM is combined to classify the fault. Firstly, the fault current signals are decomposed into certain IMF; secondly, an integral of selected IMF components along time axis is calculated to obtain the IMF energy moment eigenvectors; finally, the IMF energy moments of high correlation coefficient are taken as the eigenvectors to input into SVM classifier for fault selection. As a result, the fault selection model is obtained. The intelligent fault location system is discussed in respect of clock synchronization technology, digitizing collection technology and cloud computing. Finally based on MATLAB platform, a graphical user interface is programed and interactive operation with customer is achieved.By comparing EEMD to Fourier transform, wavelet transform and the EMD algorithm respectively, EEMD decomposition is illustrated superiority on the time-frequency analysis; The performance of three classifiers is compared and simulation data shows that the LS-SVM classification is better than the other; A simulation results of 10 kV line model shows that the proposed location method can recognize fault line accurately and effectively; GUI is tested and results verify the correctness of the software.