关键词: 电力系统, 故障诊断, 自适应搜索, 萤火虫算法, 模糊理论, 改进神经网络, 贝叶斯正则化, 粒子群

Abstract: In view of the problems existing in power transformer fault diagnosis, such as the code boundary is too absolute and the accuracy is low, a new power transformer fault diagnosis method is proposed, which applies the self-adaptive search improved glowworm swarm optimization(IGSO) to optimize the LM neural network. The method adopts the firefly individuals as the neural network’s weights and thresholds and the mean square error function of neural network as the individuals’ fitness function, and uses the IGSO to obtain the optimal weights and thresholds of the LM neural network. In the meantime, the fuzzy theory is used to handle the boundary of the improved three ratio method, and the obtained characteristic gas ratio code is used as the network model input, which has the advantages to remove the redundant information and overcome the absoluteness of code boundary. In the end, the LM neural network model of GSO algorithm is established based on self-adaptive search, and the transformer fault data is inputted for simulation. A comparison of the simulation results of the Bias regularization neural network model and the particle swarm model shows that the method has good classification performance with an accuracy rate of 88.57%.

Key words: power system, fault diagnosis, self-adaptive search theory, glowworm swarm algorithm, fuzzy theory, the improved neural network, bayesian regularization algorithm, article swarm optimization