摘要
本文给出了基于优化遗传算法的小波神经网络故障诊断模型。首先利用改进的遗传算法对神经网络的权值和阈值进行遗传操作,获得具有一定遍历性的初始权值和阈值,然后再利用神经网络的L-M训练方法进行训练,克服了BP神经网络搜索速度慢和容易陷入局部极值的缺点,保证了训练过程收敛,而且故障识别的能力和精度也大大提高。同时引进比小波分析具有更强高频分析能力的小波包技术,并将其应用到故障信号的特征频率分析中,以得到的结果作为改进遗传神经网络的输入信号,保证训练网络的准确性。通过对电机故障进行仿真试验,证实该方法的有效性及正确性。
Improved wavelet neural network based on genetic algorithm and its application in fault diagnosis of motor was presented in this paper. By introducing advanced genetic algorithm, ergodie initial values of weights and biases of neural network were researched for further net-training based on L-M algorithm. This model solved the problems of BP algorithm such as slow speed in training and liability to get into local minimum with a strong global random searching ability. It guaranteed better convergence property and ultimately improved fault diagnosing ability and accuracy. Moreover, based on wavelet packet which has more powerful analysis ability in high frequency domainthan wavelet transform, the high quality of fault-characteristic frequency vectors were obtained. With this training data for fault diagnosing network, the performance of the net was improved. Finally, the simulation result in motor indicated the high diagnosing accuracy and effectiveness of the presented net.
出处
《电子测量与仪器学报》
CSCD
2009年第3期81-86,共6页
Journal of Electronic Measurement and Instrumentation
关键词
遗传算法
神经网络
小波包
故障诊断
电机
genetic algorithm
neural network
wavelet packet transform
fault diagnosis
motor
