改进的小波神经网络在变压器故障诊断中的应用

摘要

为提高变压器油溶解气体分析法的故障诊断能力,以变压器油溶解气体作为研究对象,提出了加动量批处理小波神经网络算法.选取200组油溶解气体含量作为故障识别样本,通过多输入/多输出模式小波神经网络模型的构造,对训练过程和仿真结果进行对比分析.实验结果表明,改进的小波神经网络算法故障检测符合率高达95％,较传统的检测算法提升十几个百分点,从而极大的提高了故障诊断效率,实用性较好.%In order to improve the ability of fault diagnosis for analyzing the dissolved gases in transformer oil,a increased momentum batch wavelet neural network (WNN)algorithm was presented taking dissolved gases in transformer oil as the research objects.After the faults are recognized from 200 practical gas data,comparison and analysis are carried out in training process and simulation results with multiple-input/multiple-output-mode WNN model structure.Experimental results show that the improved algorithm fault detecting coincidence rate reaches as high as 95 ％,up to a dozen percentage points than traditional detection algorithm,greatly improving the efficiency of fault diagnosis.The algorithm has better usability.