基于BP神经网络的轨道不平顺与车体振动关联模型  

Correlation Model of Track Irregularity and Carbody Vibration Based on BP Neural Network

作　　者：牛留斌[1] 

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

出　　处：《中国铁道科学》2016年第2期26-32,共7页China Railway Science

基　　金：中国铁道科学研究院行业服务技术创新项目(2014YJ056;2014YJ089;2013YJ070)

摘　　要：利用Welch谱分析方法得到实测车体振动加速度功率谱的分布,确定车体振动能量主要集中的频段;用相干函数确定在这些频段中与车体振动相干性较强的轨道不平顺,以此轨道不平顺和对应的车辆运行速度为模型的输入变量,车体振动加速度作为模型的输出变量;考虑轨道不平顺与车体振动加速度之间存在的时间延迟步,采用主成分分析法对模型多维输入进行降维处理,构建车体振动与轨道不平顺之间3层BP神经网络关联模型,关联模型的参数由实测数据训练神经网络得到。对比模型输出与实测振动的结果表明:模型输出结果与实测数据相关程度高,波形吻合好,两者之间的残差近似符合零均值正态分布。The power spectrum distribution of the measured carbody vibration acceleration was figured out by the Welch spectral analysis method,and the frequencies that carbody vibration energy concentrated most were determined.The track irregularity of higher coherence with carbody vibration in these frequencies was selected by coherence function.The selected track irregularity and the corresponding running speed of vehicle were regarded as the input variables of the model,and the vibration acceleration of carbody was used as output variable.Considering the time delay step between track irregularity and the vibration acceleration of carbody,principal component analysis was adopted to reduce the dimensions for the multidimension input of the model.The correlation model with three-layer BP neural network was constructed for carbody vibration and track irregularity.The parameters of correlation model were obtained through training the neural network by measured data.The comparison results of model outputs and measured vibration show that model outputs have high correlation degree and better waveform agreement with measured data.The residual between them obeys approximately zero mean normal distribution.

关 键 词：轨道不平顺 车体振动加速度 BP神经网络 主成分分析 相干分析 

分 类 号：U211.5[交通运输工程—道路与铁道工程] U260.111[机械工程—车辆工程]