Transformer Fault Diagnosis Based On Improved Neural Network

Transformer is an important and expensive machine equipment in the power system,which participates in every link of power supply.If the transformer fails,it will destroy the stability of the power system and even cause power supply accidents.This paper analyzes and studies the power transformer fault diagnosis technology based on intelligent diagnosis and real-time monitoring,so as to monitor the potential fault in the transformer work in real time,improve the maintenance level of power transformer,and provide guarantee for the stable operation of the power system.For transformer fault diagnosis schemes are mainly decision tree,wavelet analysis,Bayesian classifier,neural network,etc.,and after continuous research and update,neural network diagnosis method has derived many kinds of diagnostic algorithms and optimization schemes,such as probability neural network PNN,particle swarm optimization PSO and so on.In this paper,according to the characteristic gas in transformer oil and the corresponding fault types,a transformer fault diagnosis method based on particle filter improved particle swarm optimization BP network is proposed.Firstly,the data samples to be processed are obtained by gas in oil analysis.Then,BP neural network is used for fault diagnosis.In view of the shortcomings of BP algorithm,such as slow convergence,no feedback and easy to fall into local optimal,PSO algorithm is used to optimize its structure.To solve the problems of PSO algorithm,such as too large error and precocity,a new optimization algorithm,PFPSO,was proposed by combining PSO algorithm with PF algorithm.Improved algorithm first using particle filter algorithm for oil gas data structures,hidden markov models,found in a large number of oil gas samples,a set of hidden states can through a set of data instead of the original multifarious training samples as the input of PSO algorithm,to avoid the loss caused by sample measurement error algorithm precision and performance degradation;Then,a resampling operation is added to the PSO algorithm to screen out the backward particles according to the adaptive value and modify them to enhance the PSO optimization ability.At the same time,the crossover operator and mutation operator of genetic algorithm were added in the resampling process to increase the diversity of particles and prevent the scarcity of particles.In order to ensure the search speed in the early stage and the convergence precision in the later stage,an adaptive computing method is proposed to solve the problem that the value of the crossover operator is not determined.Combined with the above,the improved fault diagnosis algorithm PFPSO-BP proposed in this paper and the traditional BP neural network algorithm,PNN probabilistic neural network algorithm,standard PSO-BP algorithm were simulated and compared by MATLAB,and the convergence algebra and diagnostic accuracy were significantly improved.