摘要
针对现有BP网络在汽车电控汽油机故障诊断中存在的问题,提出将小波函数与神经网络结合构成小波网络,代替BP网络用于故障诊断。并对小波神经网络提出了两个方面的改进。首先是对输出层函数进行了改进,其次是用熵函数代替均方误差函数作为网络的代价函数。仿真结果表明此改进的小波神经网络算法进行汽车电控汽油机的故障是有效的,而且与传统的BP神经网络相比,该改进的小波神经网络具有更强的逼近能力,更快的网络学习收敛速度和能有效避免局部最小值问题。
A new neural network-wavelet neural network is proposed for the fault diagnosis of electrical controlled gasoline engine of motorcar instead of BP network so as to improve the performance of fault diagnosis systems based on BP network. Two ways are presented to improve the wavelet network. One is the improvement of output layer function, and another is taking entropy function as the cost function of the wavelet network instead of MSE function. Simulation results indicate that this improved wavelet network is feasible for mechanical fault diagnosis. Compared with the conventional BP neural network model, this improved wavelet network has the stronger approaching capacity, the faster network-learning oonstringent velocity and can avoid local minimum issue effectively.
出处
《山西电子技术》
2008年第6期90-92,共3页
Shanxi Electronic Technology
基金
甘肃省自然科学基金项目(3ZS042-B25-039)
兰州市科技攻关项目(20816)
关键词
小波神经网络
熵函数
故障诊断
汽车电控汽油机
wavelet network
entropy function
fault diagnosis
electrical controlled gasoline engine of motorcar
