中国电机工程学报2026，Vol.46Issue(17)：6193-6205,中插5,14.DOI:10.13334/j.0258-8013.pcsee.251939
基于数据-物理模型融合驱动的城轨列车信息在线感知方法研究
Online Sensing Method of Urban Rail Transit Train Information Driven by the Fusion of Data and Physical Models
摘要
Abstract
As a typical application of intelligent transportation in urban rail transit,smart urban rail transit(SURT)enhances system efficiency through data-driven transformation with its technical foundation rooted in the synchronous perception of train position and electrical parameters.Traditional train information perception methods,highly dependent on direct measurements from on-board sensors,face challenges such as high hardware costs,difficulty in retrofitting existing trains,and poor synchronization in communicating independent position and electrical parameter data;thus,this paper proposes a data-physics model fusion-driven online perception method for urban rail train information to address these issues.The method constructs a cascaded random forest-neural network model under a collaborative data-driven and model-driven framework:first,a random forest classifier is leveraged to identify train operation organization in real time using the measurement features of traction substation(e.g.,hourly traffic rate),and then a neural network model maps the electrical parameters and position of the lowest-voltage train from traction substation data(e.g.,voltage,power).Further integrating the spatiotemporal rigid constraints of train operation diagrams,the study designs a confidence interval projection method to correct positioning results by fusing the joint confidence interval of voltage-position with the operation diagram time-axis,enabling online derivation of global train operation states.Finally,the applicability of the proposed method in long-distance power supply sections is validated through a single traction substation fault scenario from a domestic metro line.Simulation results show that under peak-hour operation density(20 train pairs per hour),the real-time global train state perception in the power supply section is effective and accurate,providing a low-cost perception solution for real-time energy management of traction power supply systems compared with traditional direct sensor measurement.关键词
直流牵引供电系统/融合驱动/信息实时感知/随机森林-神经网络级联模型/列车运行图Key words
DC traction power supply system/fusion-driven approach/real-time information perception/cascade model of random forest-neural network/train operation schedule分类
信息技术与安全科学引用本文复制引用
叶子晗,包维瀚,李笑倩,李占赫,陆超,张凌云..基于数据-物理模型融合驱动的城轨列车信息在线感知方法研究[J].中国电机工程学报,2026,46(17):6193-6205,中插5,14.基金项目
北京市自然科学基金-丰台轨道交通前沿研究联合基金项目(L221002) （L221002）
中国铁道科学研究院集团有限公司科研开发基金项目(2022YJ037) （2022YJ037）
铁科院(北京)工程咨询有限公司科研开发基金项目(2023ZX0001). Beijing Natural Science Foundation-Fengtai Joint Fund for Frontier Research on Rail Transit Project(L221002) （北京）
China Academy of Railway Science Co.,Ltd.Research and Development Foundation(2022YJ037) （2022YJ037）
Beijing Engineering consultation Co.,Ltd.of CARS Research and Development Foundation(2023ZX0001). （2023ZX0001）
