基于小波神经网络的三相整流电路的故障诊断

摘要

Some problems such as low convergence rate, small searching space and oscillation are existed in the fault diagnosis of three-phase rectifier by using neural network, an improved wavelet neural network algorithm for fault diagnosis of the thyristor of three-phase rectifier is proposed in which the momentum coefficient and alter-learning coefficient are used to resolve above problems. First, according to different output waveforms caused by different faults of thyristor, using the Multisim software to simulate the faults of three-phase rectifier, then training a modified neural network with sampling data of mal-functioning waveforms, and adopting a well trained neural network to diagnose the malfunction. The simulation demonstrates that the proposed method can provide higher diagnostic precision and require less convergence time than existing methods.%应用带动量项和自适应学习率的小波神经网络解决了应用神经网络诊断三相整流电路时收敛速度慢,搜索空闻局部极小,易引起振荡等问题.首先根据不同晶闸管的故障输出波形的不同,使用Multisim软件对三相整流电路的故障进行仿真模拟,然后用波形采集数据制作的样本对网络进行训练,最后训练好的网络可用于故障诊断.仿真表明,提出的方法比现有方法的收敛速度快,诊断误差小.