The Research Of Fault Diagnosis And Prediction For Power Transformer Based On RVM And Combinatorial Optimization Algorithm

The dependence of Chinese economic development on power is gradually deepening and the proportion of electrical energy in terminal energy consumption is increasing, a steady and reliable power supply system has became one of the important cornerstone of social economy. The power transformer is the most basic and crucial electrical equipment in electric power system, its reliability affects the safe and reliability operation of the whole power system directly, consequently, improving the operating reliability of the transformer has vevy important realistic significance. As the transformer condition-based maintenance strategy has gotten extensive application in electric network, the predicted research of early fault and diagnosis after failure for power transformer also is becoming particularly concernful.Dissolved Gas Analysis (DGA) is just a method to analyze the composition and content of some gas dissolved in transformer oil qualitatively and quantificationally, find out the cause of the gas production recur to the information of transformer oil and diagnose the internal of transformer whether running normally in order to discover the latent within fault in time. Stick out a mile, the composition and content of gas dissolved in transformer oil are the important basis and foundation for transformer fault diagnosis. Usually, transformer fault prediction is also focused on the prediction for content of the dissolved gas.The relevance vector machine (RVM) is a kind of new machine learning method that its training is based on the Bayesian framework and would get a sparse model after some not related points removed. Compared with the traditional support vector machine (SVM), RVM has many excellence such as its nuclear function does not need to meet the Mecer conditions,in addition,it has simple learning algorithm, faster operate speed, higher forecast precision and so on. This paper put forward a new transformer fault prediction method based on relevance vector machine regression prediction algorithm, the datas of the dissolved gas will train to get the relevance vector and weights by relevance vector machine, then we could make a regression prediction make use of the relevance vector and weights. The example shows that this method has favorable prediction.The characteristic of particle swarm optimization(PSO) algorithm is it has fast early convergent speed but easy to fail into local optimal solution. The cause of this phenomenon is all of the particles has higher diversity and the fitness value has bigger change in early iteration. The diversity of the community stopped in a smaller scope along with the increase of number of iterations. The specific performance is the change of fitness value become slowly and lead to premature convergence.Differential evolution (DE) algorithm choose "greedy" as its search thought in the selection process and make a comparison between individual which has mutated and intercrossed and its father generation individual, if and only if the fitness of the individual is better than father generation, make it as the next generation begain to the next iteration or keep the father generation as the next generation. The convergent speed of DE is faster, but DE also has the shortcomings of stagnation that it is easy to fall into the local advantages.According to the problems of the two algorithms, this paper put forward a new optimization algorithm combine the improved particle swarm optimization with DE algorithm.The new algorithm combine with the Fuzzy Clustering are applied to transfoemer fault diagnosis, the practical examples though the simulation test shows that this combinatorial optimization algorithm with high accuracy and fast convergent speed and has better practical guiding value for power transfoemer fault diagnosis.