基于组合预测模型的轮轨力连续测试  

Wheel-Rail Force Continuous Measurement Based on Combinational Forecast Model

作　　者：李奕璠 林建辉[1] 刘建新[1] 

机构地区：[1]西南交通大学牵引动力国家重点实验室,四川成都610031

出　　处：《西南交通大学学报》2012年第4期597-604,共8页Journal of Southwest Jiaotong University

基　　金：国家863计划资助项目(2009AA11Z202);国家自然科学基金资助项目(51075340);第12届高等院校青年教师基金基础性研究课题(121075)

摘　　要：为了精确判断车辆的运行状态,提出了一种轮轨力连续测试方法.根据轮轨相互作用的特点,采用阈值判断法从测试数据中提取轮轨力的有效信息.针对轮轨力测试系统的时变性和不确定性,将动态测试序列作为灰色过程处理,提出用灰色理论对轮轨力进行连续测试.为了提高预测精度,结合遗传算法和神经网络对传统的GM(1,1)模型进行改进.建立了10个预测模型分别进行预测,然后将精度较高的预测值输入串联灰色神经网络进行二次预测,以提高预测精度与稳定性.将这10个预测模型应用到轮轨力连续测试中,结果表明:灰色系统、遗传算法与神经网络三者的组合模型具有较高的精度,平均相对误差不超过2%,满足轮轨力连续测试的要求,并且能够降低传感器失效对测试结果的影响.In order to accurately judge vehicle operation state, a wheel-rail force continuous measure method was put forward. Based on the wheel-rail interaction characteristics, the available data of wheel-rail force were extracted from test data by using the threshold value judgmental method. From the uncertainty and time variation of a wheel-rail force measurement system, a dynamic measurement sequence was regarded as a grey process, so the grey theory was used to continuously measure wheel- rail force. In order to improve prediction accuracy, the traditional GM （ 1,1 ） model was improved by combining the genetic algorithm and neural network. Ten prediction models were established to predict respectively, and then the predicted values with high accuracy were imported into series grey neural network to predict once again to improve the prediction accuracy and stability. The 10 prediction models were applied to wheel-rail force continuous measurement. The results show that the combination model based on the grey system, genetic algorithm and neural network has a high accuracy, and the average relative error is less than 2%. This combination forecast model can meet the requirement of wheel-rail force continuous measurement and reduce the influence of sensor failures on measurement results.

关 键 词：轮轨力 车轮失圆 灰色模型 遗传算法 神经网络 组合预测 

分 类 号：U211.5[交通运输工程—道路与铁道工程]