改进PSO优化模糊RBF神经网络的牵引变压器故障诊断研究

Research on Traction Transformer Fault Diagnosis Based on Improved Particle Swarm Optimized Fuzzy Neural Network

为了快速准确地识别出油浸式牵引变压器的常见故障,结合粒子群算法全局搜索能力强、寻优速度快及模糊神经网络容错能力强、自适应性强的特点,提出了将模糊逻辑、RBF神经网络及粒子群算法有机结合的油浸式牵引变压器故障诊断方法。针对粒子群算法局部搜索能力不足的缺点,改进粒子群的速度更新公式和惯性权重,以此优化模糊神经网络的结构参数,从而构建起基于油中气体分析技术的改进PSO优化模糊神经网络的牵引变压器故障诊断模型。实验及仿真结果表明,与BP神经网络、标准PSO优化模糊神经网络相比,改进PSO优化模糊神经网络的故障辨识准确性更高、泛化能力更强。

In order to identify common faults of oil immersed type traction transformer quickly and accurate-ly ,a new integrated algorithm based on fuzzy logic, RBF neural network and particle swarm is proposed for fault diagnosis of traction transformer. Particle swarm algorithm has the global search ability and a fast search speed, and fuzzy neural network has the fault tolerant ability, strong adaptability. However,the local search ability of particle swarm is slow. Thus, improved velocity updated strategy and inertia weight strategy are proposed to optimize the structure parameters of fuzzy neural network. Furthermore, an improved particle swarm optimized fuzzy neural network based on oil gas analysis technique is formed. Experimental and simu-lation results show that the proposed method has a higher fault identification accuracy and stronger generali-zation ability comparing with BP neural network and the standard PSO optimization of fuzzy neural network.