Study Of Earth Fault Detection In Small Current Grounding Network Based On Multi-criteria Fusion Strategy

In the distribution network, small current grounding system is widely used. The existing different grounding fault line selection methods in ineffectively grounding networks collect different fault information to deal with. Since the complication of the ineffectively grounding networks and the characteristics of single-phase grounding fault, each of the methods has some limitations and shortcomings. Therefore in the paper comprehensively various methods method of earth fault detection in small current grounding system based on information fusion and intelligent process is proposed.Single phase to ground fault feature of small current grounding system is systematically analyzed. Steady-state fundamental component, steady-state fifth harmonic component, transient energy component and transient direction component extracted from zero-sequence current separately by means of FFT and wavelet packet transforms.Based on the analysis of a variety of line selection methods the advantages of a variety of fault line selection methods is combined by fuzzy theory to improve the accuracy of fault selection line. Due to fuzzy characteristics of fault, first of all, the various components of the membership function is constructed, and then different weights coefficient is determined according to the application of each method to carry out fuzzy comprehensive evaluation for fusion result. In order to make results more reliable least squares optimization is used for the weight coefficient matrix. At last two types of Neral Network-Fuzzy Systems combining fuzzy systems and neural networks which have excellent self-learning capability is used for multiple fault selection line criterias fusion.The structure and training algorithm of Neral Network-Fuzzy Systems is designed. The input and output of Neral Network-Fuzzy Systems are fuzzed and the algorithm is modified. Fault samples are used for train and test of Neral Network-Fuzzy Systems.The distribution network model built in Matlab7.1 to simulation for fault data in fault ground types, fault location, fault close initial angles, different lines to demonstrate method feasibility.