Research On A New Method To Detect The Contamination Condition Of Insulator Based On Fuzzy Neural Network

With the great development of national economy and the growing of Power System Voltage class, contamination flashover of insulators is having more minus effect on national economy.Theoretical analysis along with laboratory study has shown that the variety of leakage current is closely related to its running state. But how to evaluate the state of insulator with the information collected by the On-line monitoring system is still a problem. On the basis of searching and studying plenty of internal and oversea reference literature and research findings, a new method to detect the condition of insulator based on Fuzzy Neural Network is present .Using this method we can get real-time value of insulators state in fixed quality, which improves the detecting veracity and the ability of predicting insulator flashover and makes a significant change from plan-based maintenance to condition-based maintenance.Firstly the mechanism of contamination flashover is researched in the paper, and the characteristic of leakage current and environment factors that influence leakage current are analyzed. Then the insulator detecting model based on Fuzzy Neural Network is put forward .In the model, the filtered field date collected by On-line monitor was calculated by fuzzy membership function, then it is trained and tested as ANN (Artificial Neural Net) input using MATLAB Neural Network Tools. The test result shows that it can predict the state of insulator correctly. By this way it can prewarn or give an alarm on insulators' contamination and humidity state, so the personnel can get the real-time state on insulator surface to determine whether it needs cleaning. An active partial nonperiodic cleaning method will substitute the former passive general fixed cleaning method in order to prevent contamination flashover.It lays emphasis on determinations of fuzzy neural network model and its characteristic quantity of input and output and its membership function. The number of hide-layer is confirmed by observing its influence on the convergence of network. To avoid getting into local optimum in BP arithmetic, an improved BP arithmetic--additonal momentum method is adopted to train fuzzy neural net. The result of simulation showed that it had great accuracy in insulators status detection. You don't have to determine the value of weights because it can automatically determine weights and threshhold values after self-learning training on field samples, which can be used to access contamination condition by inputing other samples besides train set. BP net has high self-adaption ability, high calculation accuracy. Its evaluation results are more objective and accurate compared with insulators fuzzy logic evaluation method.