自适应GA优化WNN的模拟电路软故障诊断方法*

在小波神经网络(WNN)的模拟电路故障诊断系统中,普遍采用的梯度下降算法在训练时易使网络陷入局部最优,而网络结构的冗余也会造成训练收敛方向偏离全局最优点,降低推广能力和增加误诊率。用自适应遗传算法优化WNN,以克服上述缺陷。采用该方法可简化小波神经网络的结构和优化参数,在滤波器电路的软故障识别中获得满意的效果。与常规的WNN故障诊断方法相比,有效地提高了故障诊断的效率和正确率。

Adaptive genetic algorithm optimized WNN approach for analog circuit soft fault diagnosis

In analog circuit fault diagnosis system using wavelet neural networks (WNN), the prevalent algorithm, gradient descent algorithm, is prone to make WNN converge to the local minimum in training phase. Additionally, the structure redundancy of network may lead to training convergence direction deviating from globally optimal point so that the network gegenerality will be degraded and diagnosis inaccuracy increased.This paper proposed the adaptive genetic algorithm for optimizing WNN to avoid the limitation above. This approach could achieve simplified structure and optimized parameters for WNN, which obtained satisfactory effects in soft fault identification for filter circuit. The presented method gained better diagnosis efficiency and accuracy in comparison with conventional WNN approach.