基于小波分析和遗传神经网络的模拟电路故障诊断方法

摘要

针对BP网络极易收敛于局部极小点与过拟合等缺点,在构建小波神经网络的基础上,提出用遗传算法优化BP神经网络的模拟电路故障诊断方法.该方法使用小波作为预处理工具,经PCA分析和归一化后提取输出信号的能量信息作为特征向量,用遗传BP神经网络作为故障识别器,对模拟电路故障进行诊断.与传统BP神经网络相比较,结果表明,该方法可明显改善神经网络结构、提高故障诊断的精度和速度.%In order to solve the problems of BP( back-propagation) network usually converges to local minimum and over-fit-ting and other shortcomings, this paper developed a wavelet decomposition and GA based approach for analog circuits. Using the wavelet decomposition as a preprocessor, extracted the feature information by wavelet de-noising and optimized BP by GA. A comparison of our work with BPNN, which reveals that this work improves network structure and increase fault diagnosis preci-sion and velocity.