摘要
针对多相整流电路的特点,采用对输出电压分段平均值进行谱分析的方法,得到网络的特征量,不但降低了网络样本的维数,而且使得特征量与采样的误差和整流控制角的相关性减到最小;导出一种既有最速下降法的稳定性又有牛顿法的快速性的新算法;设计了该算法的神经网络结构,并讨论了网络结构参数对故障诊断结果的影响;此方法简单易行,误判率低,训练时间短,对各种多重多相化的整流电路具有普遍的适用性.
According to the characteristics of the multi-phase rectifier, the eigenvalue of network is obtained form the spectrum analysis of the sub-area voltage mean value. One of the advantage is the decrease of the sampling data dimension. The correlation between the eigenvalue with simple error and triggering angle are eliminated. A improved BP neural network algorithm with the stability of the steepest descent method and the speed performance of the Newton approach method is educed. The structure of the neural network is designed and the influence caused by the network parameters is discussed. This arithmetic processes the low misdiagnosis rate and short training timing. It has universal applicability for other kinds of the multi-phase rectifier.
出处
《扬州大学学报（自然科学版）》
CAS
CSCD
2006年第1期53-57,共5页
Journal of Yangzhou University：Natural Science Edition
基金
扬州大学科研基金资助项目(U0211096)
关键词
多相整流电路
故障诊断
神经网络
multiple rectifying circuit
fault diagnosis
neural network
