EPS助力特性的改进RBF神经网络方法  

An Improved Method of RBF Neural Network for Power-Assisted Characteristic of the EPS

作　　者：文科[1] 那文波[1] 杨惠忠[1] 

机构地区：[1]中国计量学院机电学院,杭州310018

出　　处：《机械设计与制造》2012年第8期69-71,共3页Machinery Design & Manufacture

基　　金：浙江省重大科技专项(2008C01038-2)

摘　　要：针对电动助力转向(EPS)助力特性的非线性问题,提出应用径向基(RBF)神经网络强非线性能力进行电动助力转向(EPS)助力特性曲面拟合方法,并做出改进。应用改进均值聚类方法(k-means)对数据进行聚类,获取基函数参数,再用梯度下降法训练网络权值,并利用最优停止法对网络进行了优化。实验结果表明,该改进方法避免了过拟合现象,提高了网络的泛化能力,并且具有网络训练时间短,拟合的曲面精度高,预测能力强等优点。It presents a method to fit the power-assisted characteristic curve of the electric power steering（EPS）based on nonlinearity of RBF neural network,and make improvements to the RBF neural network.An improved k-means clustering algorithm is used to cluster the data and the radial basis function parameters are obtained,while weights of the network are trained with gradient descent approach and the network is optimized through the optimal stopping rule.Experiments and simulations show that the method avoids overfitting phenomenon and enhance the network generalization ability with short network training time,high precision in curve fitting and strong in prediction ability.

关 键 词：电动助力转向(EPS) 助力特性曲面 径向基神经网络(RBFNN) 最优停止法 

分 类 号：TH16[机械工程—机械制造及自动化] U463.4[机械工程—车辆工程]