Research On Transformer Fault Detection Technology Based On Improved PSO-RBF Neural Network

Transformers are vital to the power grid system and are the most vulnerable type of equipment.If the transformer fault can be detected in time and the cause can be discovered,the major losses and impacts caused by the damage of the transformer can be avoided.Combining photoacoustic spectral gas detection technology with intelligent fault diagnosis models enables online monitoring of transformers.Radial basis function(RBF)neural network has fast convergence speed,simple structure,and strong cluster analysis ability,which is very suitable for classification problems.The particle swarm optimization(PSO)algorithm is a swarm intelligence optimization algorithm with fast convergence speed,fewer parameters to be controlled,as well as excellent robustness and usability.It has been gradually applied to the parameter determination of neural network.This paper designs a new model for optimizing the RBF neural network using an improved PSO algorithm and applies it to the photoacoustic spectroscopy detection system.The main content of this paper is as follows:(1)Analyze the causes of transformer faults,summarize the types of transformer faults and corresponding characteristic gas types,and expound methods based on gas analysis in oil and the deficiencies of traditional fault diagnosis methods.The related principles of RBF neural network are introduced,and the feasibility of RBF neural network in transformer fault diagnosis application is verified by MATLAB simulation.(2)Use the PSO algorithm to optimize the RBF neural network and propose improvements to the particle swarm algorithm.Through the analysis of the principle,advantages and disadvantages of the particle swarm algorithm,the PSO algorithm is improved from its two key parameters.Aiming at the inertia weight,this paper proposes a cycloid-type inertia weight declining model whose prototype is the fastest curve;for the learning factor,it adopts dynamic changes to improve.Four test functions were used to test the IPSO algorithm through MATLAB simulation,and compared it with the standard PSO algorithm and the linear drop inertia weight PSO algorithm to verify the performance advantage of improved PSO algorithm.(3)Establish an IPSO-RBF neural network model and conduct simulation tests,compare the IPSO-RBF neural network model with a RBF neural network trained by conventional algorithm and a standard PSO-RBF neural network,use MATLAB to model and simulate.From the three aspects of network output error,calculation time and fault diagnosis accuracy rate,it is verified that the IPSO-RBF neural network not only has high accuracy and good stability,but also its correct identification rate of the fault type is over95%,which meets the needs of practical applications.The RBF neural network model is written into the desktop software for online transformer monitoring and is actually used in the photoacoustic spectroscopy system.