基于数据特征识别的接触网典型设备状态不良预警  

Early warning of typical equipment malfunctions in overhead contact systems based on data feature recognition

作　　者：王同军 柯在田 王婧 WANG Tongjun;KE Zaitian;WANG Jing(China Railway Society,Beijing 100844,China;Infrastructure Inspection Research Institute,China Academy of Railway Sciences Group Co.,Ltd.,Beijing 100081,China)

机构地区：[1]中国铁道学会,北京100044 [2]中国铁道科学研究院集团有限公司基础设施检测研究所,北京100081

出　　处：《铁道科学与工程学报》2025年第3期1396-1406,共11页Journal of Railway Science and Engineering

基　　金：中国国家铁路集团有限公司科研项目(N2024G020)。

摘　　要：随着高速铁路运营时间持续增加,因接触网设备状态异常导致的弓网故障时有发生,为提前发现设备运行安全隐患,提出基于数据特征识别的典型设备状态不良预警方法。首先,提出接触网设备状态不良预警模型,该模型以“数据准备-特征提取-算法构建-试验验证”为主轴,给出了各典型设备状态不良预警方法构建的主要技术项点和路线;然后,采用卡尔曼滤波、角点特征识别、动态时间规整等方法,给出了接触网检测数据粗差修正算法、定位点识别算法和里程校正算法,为开展预警方法构建提供了良好的数据基础;最后,基于弓网综合检测系统的实测数据,针对接触网运行维护中常见的几何状态、定位器受力状态和弹性状态不良问题,通过多种时序数据分析方法提取了数据特征,构建了训练样本库,并采用离群点诊断、随机森林等算法训练获得了3项典型设备状态不良预警方法。对获得的各项典型设备状态不良预警方法开展试用,发现了接触网支柱倾斜、腕臂套管滑移、定位部件磨损、补偿装置卡滞等多类严重设备状态异常处所,总体准确率93%,表明本文方法能够精准定位典型设备状态异常处所,具备在接触网运维中应用的可行性。通过分析算法发现的设备状态不良处所,验证了本文提取的各项数据特征可分性良好,能够准确反映接触网设备状态变化。With the continuous increase in the operation time of high-speed railways,pantograph-catenary failures caused by abnormal conditions of OCS equipment occur from time to time.To identify potential safety hazards in equipment operation in advance,a typical equipment condition deterioration early warning method based on data feature recognition was proposed.Firstly,a OCS equipment condition deterioration early warning model was proposed.This model was centered around the“data preparation-feature extraction-algorithm construction-experimental verification”axis,providing the main technical points and routes for constructing an early warning method for equipment condition deterioration.Then,using methods such as Kalman filtering,corner feature recognition,and dynamic time warping,algorithms for gross error correction of OCS detection data,positioning point recognition,and mileage correction were presented,providing a solid data foundation for the construction of early warning methods.Finally,based on the actual measurement data from the pantographcatenary comprehensive detection system,for common geometric conditions,force conditions,and elasticity conditions in OCS operation and maintenance,data features were extracted through various time series data analysis methods.A training sample library was constructed,and three typical equipment condition deterioration early warning methods were obtained through training using outlier diagnosis,random forest,and other algorithms.The obtained typical equipment condition deterioration early warning methods are tested,and various serious equipment condition abnormalities such as mast tilt,cantilever sleeve slippage,positioning component wear,and compensation device jamming are discovered.The overall accuracy rate is 93%,indicating that the proposed method can accurately locate typical equipment condition abnormalities and is feasible for application in OCS operation and maintenance.By analyzing the equipment condition deterioration locations discovered by the algorithm,it is

关 键 词：OCS 状态预警 特征建模 机器学习 数据预处理 

分 类 号：U225.4[交通运输工程—道路与铁道工程]