基于粒子群优化神经网络的变压器故障诊断  

Power Transformer Fault Diagnosis Based on Neural Network Evolved by Particle Swarm Optimization

作　　者：王晓霞[1] 王涛[2] 

机构地区：[1]华北电力大学计算机科学与技术学院,保定071003 [2]华北电力大学数理学院,保定071003

出　　处：《高电压技术》2008年第11期2362-2367,共6页High Voltage Engineering

基　　金：国家自然科学基金(10774043)~~

摘　　要：为克服电气分析应用中误差反向传播(BP)神经网络存在的不足,提出了一种利用改进粒子群算法优化神经网络的变压器故障诊断新方法。该法的惯性权重自适应调整,以平衡局部和全局搜索能力;收缩因子加快算法的收敛速度,有利于更快地收敛于全局最优解。利用改进的粒子群算法优化神经网络参数,并结合BP算法训练网络可有效地克服常规BP算法训练网络权值和阈值收敛速度慢、易陷入局部极小和遗传算法独立训练神经网络速度缓慢等缺点。最后,进行变压器故障实例分析的仿真结果表明,该算法具有较快的收敛速度和较高的诊断准确度,证实了该方法的正确性和有效性。Power transformer is one of the most important equipments in power network. Its normal operation is the basis of the power supply and normal social life. So it is valuable to discover the incipient fault accurately and timely. This paper proposes a new power transformer fault diagnostic method using neural network evolved by modified particle swarm optimization （modified PSO） algorithm in order to overcome the problem of premature convergence observed in many applications of error back propagation （BP） algorithm and enhance the fault diagnostic ability of conventional dissolved gas-in-oil analysis in power transformer. In the modified PSO algorithm, the inertia weight is adjusted adaptively in order to balance and reconcile the global and local searching capability. The convergence can be accelerated by setting the compression factor of the modified PSO algorithm reasonably, which may benefit to find the global optimal solution quickly. Firstly, the modified PSO algorithm is used to optimize the original parameter of the neural network, then the gradient descent algorithm is used to train the neural network. Defects of conventional BP algorithm, i.e. the slow convergence of weight and threshold learning, premature result, and the slow training speed of GA, are settled by the algorithm. Finally, Simulation results of power transformer fault diagnosis show that both convergence speed and diagnosis accuracy are improved to some extent. That shows the correctness and validity of this method in power transformer fault diagnosis by dissolved gas-in-oil analysis.

关 键 词：粒子群优化算法 BP算法 神经网络 变压器 故障诊断 仿真 

分 类 号：TM41[电气工程—电器] TP18[自动化与计算机技术—控制理论与控制工程]