摘要
针对检测车对ZPW-2000系列无绝缘轨道电路(JTC)的健康评价能力需要提升的实际需求,基于机器学习理论提出一种JTC健康状态综合评价方法。首先,基于JTC的工作原理和检测车的数据分析,研究并确定JTC健康状态评价指标和评价函数。然后,基于JTC仿真模型,使用故障注入技术对JTC常见故障模式下的数据进行仿真,构建JTC状态数据集,并基于半监督聚类算法,结合传统层次分析法的专家经验和数据中的规律,实现数据标注。最后,基于XGBoost模型和SHAP可解释框架计算各评价指标的权重,进而构造JTC健康分数HI,实现JTC健康状态的综合评价。实验表明,本文方法能够对JTC进行准确、合理地评价,克服现有评价方法的不足,有效提升检测车对JTC的健康评价能力,为实现“状态修”提供依据。
In response to the actual demand for improving the health evaluation performance of the inspection train for the ZPW-2000 jointless track circuit(JTC),this paper proposed a method of comprehensive health evaluation of JTC based on machine learning theory.Firstly,the evaluation indexes and evaluation functions of JTC health status were studied and determined based on the working principle of JTC and data analysis of the inspection train.Then,based on the JTC simulation model,fault injection technology was used to simulate JTC data in each common fault mode and construct the JTC status dataset.The expert experience of traditional analytic hierarchy process(AHP)and the rules in data were used for data annotation based on semi-supervised clustering algorithm.Finally,the weight of each evaluation index was calculated based on the XGBoost model and SHAP interpretable framework,and the JTC health index(HI)was constructed to realize the comprehensive health evaluation of JTC.The experiments show that the proposed method,which can evaluate JTC accurately and reasonably and overcome the shortcomings of existing evaluation methods,can effectively improve the health evaluation performance of the inspection train for JTC,and provide the basis for realization of condition based maintenance.
作者
董寅超
赵林海
DONG Yinchao;ZHAO Linhai(School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China)
出处
《铁道学报》
EI
CAS
CSCD
北大核心
2023年第12期92-102,共11页
Journal of the China Railway Society
基金
国家重点研发计划(2018YFB1201505-05)。
关键词
无绝缘轨道电路
检测车
健康评价
机器学习
jointless track circuit
inspection train
health evaluation
machine learning
