GA-BP网络在钢轨磨损量预测中的应用  

Application of GA-BP ANN in the Prediction of Wear Volumes of Rail Steel

作　　者：王平[1] 王彩芸[1] 王文健[1] 刘启跃[1] 

机构地区：[1]西南交通大学牵引动力国家重点实验室摩擦学研究所,四川成都610031

出　　处：《润滑与密封》2011年第2期99-102,71,共5页Lubrication Engineering

基　　金：国家科技支撑计划项目(2009BAG12A01-B05-1);国家重点基础研究发展973计划(2007CB714702);国家863计划项目(2009AA04Z424)

摘　　要：基于GA算法及BP神经网络优点,将GA算法优化的BP网络应用于钢轨磨损量预测。通过赫兹模拟试验方法,在JD-1轮轨模拟试验机上获得不同曲线半径、轴重、运行速度工况下对应的钢轨磨损量,即神经网络所需样本,建立GA算法优化三层BP神经网络,对钢轨磨损量进行预测。结果表明:GA算法优化的BP神经网络对钢轨磨损量具有良好的预测性能,较好地反映了曲线半径、轴重、运行速度对钢轨磨损量的影响规律。Based on advantages of GA and BP ANN, BP ANN optimized by GA was applied to predict rail wear volume. With the Hertz testing simulation method, the experiment investigation on rolling wear of rail-wheel was performed on JD-1 wheel/rail simulation facilities to obtain wear volume in different conditions （curve radius, axle load, velocity）. A feed-forward backpropagation network with three layers was created to predict wear volume of rail steel. The results show that BP ANN optimized by GA has good properties of prediction in wear volume of rail, which can reflect and report the effect of curve radius, axle load and velocity on wear volume of rail steel better.

关 键 词：GA BP神经网络 钢轨 磨损量 

分 类 号：U213.42[交通运输工程—道路与铁道工程]