Application of Oil-immersed Transformer Fault Diagnosis Based on Modified Neural Network Combined with PSO

Transformer fault diagnosis is a non-linear pattern recognition process. A single BP neural network has a slow convergence speed and is easy to fall into a local minimum. In this paper, based on the gas characteristics and fault types of oil-immersed transformers, a method is proposed on account of the dissolved gas analysis technology in transformer oil. A modified the transformer fault diagnosis algorithm of particle swarm optimization with neural network is suggested. By dynamically adjusting the inertia weight and acceleration factor of the PSO algorithm; by optimizing the standard particle swarm algorithm, an adaptive PSO-BP neural network transformer fault diagnosis system is established. The new and modified algorithm is used to establish a fault diagnosis model, and more than 300 samples are used for actual fault diagnosis, with an accuracy rate of over 94%. The results show that the modified PSO-BP is more accurate and reliable.