基于LVQ神经网络的轨道单元状态综合评判方法研究  

Study on comprehensive evaluation method of track unit condition based on LVQ( Learning Vector Quantization) nerve network

作　　者：许贵阳[1] 刘金朝[1] 曲建军[1] 史天运[2] 

机构地区：[1]中国铁道科学研究院基础设施检测研究所,北京100081 [2]中国铁道科学研究院电子计算技术研究所,北京100081

出　　处：《铁道建筑》2013年第10期84-87,共4页Railway Engineering

基　　金：国家自然科学基金资助项目(51178464);国家"863"计划项目(2011AA11A102);国家科技支撑计划项目(2011BAG05B02-01)

摘　　要：为了有效利用多种检测数据评判轨道单元的状态,提出利用LVQ(学习矢量量化)神经网络建立轨道单元特征参数与轨道单元分级的关联模型,通过对TQI(轨道质量指数)、轨道几何、加速度、晃车仪、添乘仪、人体感觉的超限扣分加权得到轨道单元的量化评分指标,并利用层次分析法确定各特征参数的权系数。根据大量实测数据建立随机样本,利用聚类方法确定轨道单元状态的分级。以轨道单元的量化评分指标作为输入,以聚类得到的表征轨道单元分级的矢量量化数据作为输出,利用误差反向传播方法训练LVQ神经网络模型。利用新的评判方法对某线路的轨道单元状态进行评判,结果表明该方法可行、有效,为轨道单元状态综合评判提供了一条新途径。The paper proposes to build the correlation model for the track unit under discussion, intending to study the relation between its characteristic parameters and its classification, based on Learning Vector Quantization （LVQ） neural network,so as to evaluate the state of the tract unit through a large range of data. In this case, analytic hierarchy process can be used to determine the weighted coefficient for each characteristic parameter, therefore quantified assessment index can be reached as track quality index （TQI）, track geometry, acceleration, vibration-acceleration inspection devices and passengers＇s general feeling were quantified and deducted by its weighted sum in excessive cases. Random samples were built based on a considerable amount of data drawn from measurement and the classification of the track unit was determined with the introduction of cluster analysis. The error back propagation was applied to further improve the LVQ neural network model with the quantified assessment index as input and the vector quantification data as output. The papar evaluated the track unit state with the help of the new evaluating method, which turned out to be plausible and effective, therefore provide new approach for the integrated assessment of track unit.

关 键 词：轨道单元 学习矢量量化 神经网络 层次分析法 聚类方法 

分 类 号：U213.213[交通运输工程—道路与铁道工程]