混合蛙跳算法优化神经网络的同步电机转子故障检测

摘要

By analyzing the linear relationship between the salient pole synchronous generator different rotor winding inter-turn short circuit level and the the excitation current,wavelet packet analysis and shuffled frog leaping algorithm neural network combined diagnostic method was presented in this paper.As the BP neural network input signal,the excitation current of wavelet packet decomposition and wavelet packet coefficient were refactored,the structure feature vector for each frequency band signal energy was calculated also.The improved shuffled frog leaping algorithm (ISFLA) of BP neural network was put forward.In the novel method,shuffled frog leaping algorithm (SFLA) was proposed for BP neural network's initial weights and threshold optimization,the chaos operator and contraction factor were introduced to improve the traditional shuffled frog leaping algorithm updating strategy.Matlab simulation results show that the proposed SFLA-BP neural network is able to detect the inter-turn short circuit fault levels for rotor of the salient pole synchronous machine effectively.%通过分析隐极式同步发电机不同转子绕组匝间短路程度与励磁电流之间存在的线性关系,提出利用小波包分析和混合蛙跳算法神经网络相结合的诊断方法,对励磁电流进行小波包分解和小波包系数重构,求解各频带信号能量构造特征向量,作为BP神经网络的输入信号.提出改进的混合蛙跳算法ISFLA-BP神经网络,利用混合蛙跳算法(SFLA)对BP神经网络的初始权值进行优化,并引入混沌算子和收缩因子,改进传统混合蛙跳算法的更新策略.Matlab仿真表明,本文提出的ISFLA-BP神经网络能够有效地检测隐极式同步发电机转子匝间短路故障程度.