Research On Transformer Fault Diagnosis Based On Fish Swarm Optimization And BP Neural Network

With the rapid development of the social economy,the demand for electricity is increasing,and the scale of China's power industry is growing.Achieving the safety,economy and stability of the grid will bring greater challenges to the power sector.Power transformers play an important role in the power transmission and transformation process of the power grid.Therefore,realizing fault diagnosis of transformers,especially online monitoring and diagnosis without affecting the normal operation of the power grid to ensure safe and reliable economic power transmission and transformation,has important research value.In view of the development of online monitoring technology and artificial intelligence technology and the advantages of dissolved gas analysis(DGA)technology in the field of transformer fault diagnosis,this paper uses artificial fish swarm algorithm(AFSA)to optimize BP as a fault diagnosis method.This paper first summarizes the current research status of power transformer fault diagnosis at home and abroad,and then introduces several typical fault types of transformers in detail,then analyzes the Rogers ratio method,the characteristic gas three-ratio method and the intelligent fault-based transformer fault diagnosis method based on DGA.The principle of BP and AFSA is also introduced in the theoretical part.In the experimental part,70 sets of data were collected by oil chromatography analysis,and then the structure of BP was determined according to the type of characteristic gas and the type of fault.The first 60 sets of data were trained,and the last 10 sets of data were tested during the test.BP has been found to have local convergence defects,but this problem can be solved by trial and error by adjusting the number of neurons in the hidden layer.The use of PSO,GA and AFSA need to determine the fitness function.However,the mathematical model of transformer fault diagnosis is very fuzzy.It is very difficult to write the mathematical expressions of multiple input and multiple output,that is,the fitness function cannot be determined.At the same time,considering the advantages of BP to establish a fuzzy mathematical model,the problem of determining the fitness function can be solved by the combination of the above three intelligent algorithms and BP.Select the same sample data in MATLAB for testing with four algorithms including BP,PSO optimization BP,GA optimization BP and AFSA optimization BP,and record BP training steps,BP convergence time,the number of iterations of the algorithm optimization,the optimal fitness value and the prediction accuracy rate as evaluation indicators.The comparison of the four algorithms shows that the AFSA optimized BP neural network performance is the best,and the proposed algorithm AFSA optimization BP is better to solve the transformer fault diagnosis.