摘要： 针对电厂汽轮发电机组故障诊断问题，将小波变换和自联想神经网络结合构造了一个多层的自联想小波网络故障诊断系统。在输入层对振动信号进行二进离散小波变换，提取其在多尺度下的细节系数作为故障特征向量，根据这些特征向量进行自联想网络的学习，用学习过的自联想网络诊断故障。将该方法成功地应用于汽轮发电机组故障诊断，实验仿真结果表明，该方法优于常规的BP网络方法：某些单一故障的识别率提高了31.2%，综合故障的识别率提高了26.6%。

关键词: 神经网络, 小波变换, 故障诊断, 自联想

Abstract: Multiple-layer auto-associative wavelet network is presented to solve the problem of fault diagnosis for turbo-generator units, which combines wavelet transform and Auto-Associative Neural Network(AANN). Vibration signal is processed by discrete binary wavelet transform at the input layer and the detail coefficients are obtained under multi-resolution as fault character vectors. AANN is trained according to these character vectors. The trained AANN diagnoses the fault. This method is successfully used to diagnose the fault of turbo-generator units. Simulation result proves that the method is better than the regular BP network. Single fault recognition rate rises by 31.2% and comprehensive fault recognition rate rises by 26.6%.

Key words: neural network, wavelet transform, fault diagnosis, auto-associative

中图分类号: 

• TP311