改进CLR的预测算法在铁路机车牵引系统故障维修中的应用  

Application of Improved CLR Prediction Algorithm in Fault Maintenance of Railway Locomotive Traction System

作　　者：李曼[1] 宾紫嫣 周鑫燚 覃思瑶 LI Man;BIN Ziyan;ZHOU Xinyi;QIN Siyao(State Key Lab of Rail Traffic Control&Safety,Beijing Jiaotong University,Beijing 100044,China;School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学轨道交通控制与安全国家重点实验室,北京100044 [2]北京交通大学交通运输学院,北京100044

出　　处：《铁道运输与经济》2024年第3期156-163,188,共9页Railway Transport and Economy

基　　金：国家自然科学基金青年基金项目(52002019);北京交通大学轨道交通控制与安全国家重点实验室自主研究课题(RCS2022ZT006)。

摘　　要：准确及时地预测机车系统故障是有效保障铁路运输安全与合理制定设备维护策略的关键。现有故障预测多集中于对特征数据的挖掘与分析。利用校准标签排名(CLR)结合自适应簇数的聚类算法对牵引系统故障类型进行预测,通过聚类获得故障类型深层特性,采用CLR算法从系统特征参数获取故障类型的相关性排名。通过加入人工校准标签来预测故障类型的相关性,同时减轻类不平衡问题的负面影响。相较于经典CLR算法,改进后的算法有一定幅度的性能提升。以牵引系统主变压器实际故障数据集为例,对冷凝器漏油、冷却风机异音等多种故障类型进行预测,结果表明:单个损失降低了78.8%,汉明损失提升了15.6%,故障维修方式预测准确率达96.4%,为设备故障预测性维护工作提供理论支撑。Accurate and timely prediction of locomotive system failures is the key to effectively ensuring railway transportation safety and rationally formulating equipment maintenance strategies.However,the existing fault prediction modes mostly focus on the mining and analysis of feature data.Calibration Label Ranking(CLR)combined with an adaptive clustering algorithm was proposed to predict the fault types of traction systems in this paper.The deep characteristics of the fault types were obtained through clustering,with the CLR algorithm used to obtain the correlation ranking of the fault types from the system feature parameters.The manual calibration labels were added to predict the correlation of fault types,while the negative impact of class imbalance issues was alleviated simultaneously. Compared with the classical CLR algorithm, the improved algorithm showed some improvement in performance. Taking the actual fault dataset of the main transformer of the traction system as an example, multiple types of faults such as oil leakage of the condenser and abnormal noise of cooling fans were predicted. The results show that the single loss decreased by 78.8%, the Hamming loss increased by 15.6%, and the accuracy of fault maintenance method prediction reached 96.4%, providing theoretical support for predictive maintenance of equipment failures.

关 键 词：故障维修预测 牵引系统 成对比较排序 CLR 人工校准标签 

分 类 号：U260.1[机械工程—车辆工程]