马立新　范丽君

摘要：

在电力能效监控管理系统中，提出了基于小波包的特征提取和BP（back propagation）神经网络相结合的方法，对三相整流电路中故障晶闸管位置进行诊断和识别.根据整流电路原理，对22种故障情况分别进行编码.建立三相整流电路故障模型，采用小波包分解的方法，对直流端输出电压的采样数据进行特征提取，构建特征向量，作为BP神经网络的训练样本，将对应故障的编码作为网络输出，用简化的训练好的神经网络即可以实现整流电路的故障位置识别.仿真结果证明，采用小波包特征提取，作为神经网络训练样本，既可以简化神经网络训练结构，又可以准确实现故障定位识别.研究具有很大的工程实践意义.

关键词：

电力能效测评； 小波包； 特征向量； 神经网络； 整流电路； 故障识别

中图分类号： TM 92文献标志码： A

Abstract：

In the power energy efficiency management system，the feature extraction based on wavelet packet combining with back propagation（BP） neural network was proposed and applied to thyristor fault diagnosis and identification in the threephase rectifier circuit.According to the principles of rectifier circuit，22 kinds of fault were encoded respectively.The fault model of threephase rectifier circuit was set up.Using the wavelet packet decomposition method，feature extraction of the DC output voltage was conducted to construct the feature vectors，which was saved as training samples of BP neural network.The corresponding fault codes were used as the network output.This simplified trained neural network could recognize the fault position of the rectifier circuit.The simulation results showed that the wavelet packet feature extraction，used as the neural network training sample，not only simplified the structure of neural network training，but also located the fault thyristor accurately.It indicated the engineering significance.

Keywords：

electric energy efficiency evaluation； wavelet packet； feature vector； neural network； rectifier circuit； fault identification

三相整流电路广泛应用于电气设备中.晶闸管本身损坏以及触发脉冲一场导致的不导通和误导通都会使该晶闸管所在的整流电路发生故障以至于整流电压畸变.因此，对电力电子电路实现在线实时监测和故障诊断显得很有必要.在对故障诊断快速性和准确性要求越来越高的同时，人们也不断寻找如何对三相全控整流电路中故障晶闸管快速、准确地定位，应用先进的算法实现智能故障诊断也越来越受到重视.

传统的检测方法有电压电流检测法、傅里叶分析法、频谱分析与神经网络相结合、粗糙集与神经网……