基于Volterra级数与神经网络的非线性网络的频域故障字典诊断法

A Fault-Dictionary Diagnostic Method in Frequency Domain for Nonlinear Networks Based on Volterra Series and Backward Propagation Neural Networks (BPNN)

提出了用Volterra频域核(传递函数)作为故障特征对非线性定常动态网络进行故障诊断的方法,即计算出网络响应在各种常见故障状态下的Volterra级数解的各阶频域核(3～4阶即可),并将其输入给BP神经网络(BPNN).利用BPNN的分类功能建立故障字典,对实测的故障网络的各阶频域核进行测试样本分类,实现故障诊断.给出了各阶频域核的统一递推算式,并讨论了Volterra频域核的实验测量方法以及基于人工神经网络(ANN)的求解方法.

This paper proposed a fault-dictionary diagnostic method in frequency domain for nonlinear network based on Volterra series and BPNN. We calculated the frequency cores of Volterra series from the responses of the network in various fault states, which can be taken as fault features and were fed into BPNN. By using the classifiable function of BPNN, we built a fault dictionary. After classifying the Volterra frequency cores tested from the networks, the fault location was completed. We established the unified recursive computing formulas for the frequency cores and discussed the test method for the frequency cores and the method for solving the equations by ANN.