考虑驾驶特征识别的西部地区高速列车追踪运行轨迹预测方法研究  

Research on Prediction Method of High⁃speed Train Tracking Trajectory Based on Driving Feature Recognition

作　　者：田婉琪 步兵[3] 吕继东 唐涛[3] 李开成[1,2] TIAN Wanqi;BU Bing;LYU Jidong;TANG Tao;LI Kaicheng(National Engineering Research Center of Rail Transportation Operation and Control System,Beijing Jiaotong University,Beijing 100044,China;School of Automation and Intelligence,Beijing Jiaotong University,Beijing 100044,China;State Key Laboratory of Advanced Rail Autonomous Operation,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学轨道交通运行控制系统国家工程研究中心,北京100044 [2]北京交通大学自动化与智能学院,北京100044 [3]北京交通大学先进轨道交通自主运行全国重点实验室,北京100044

出　　处：《铁道学报》2025年第1期71-81,共11页Journal of the China Railway Society

基　　金：中央高校基本科研业务费(2022JBXT000);国家自然科学基金(52272329);北京市自然科学基金(L211019)。

摘　　要：针对西部地区高速铁路列车在不同应用场景下的驾驶特征和轨迹的差异性较大的问题,建立基于驾驶特征分类的高速铁路列车追踪运行轨迹预测模型。给出高速列车驾驶特征定义,结合追踪列车的速度、加速度、急动度、时间和空间间隔等特征数据,基于高斯混合模型(GMM)识别出3类常见且差异较大的高速列车驾驶场景和驾驶风格,并对驾驶风格的最大互信息系数(MIC)特征进行重要性分析;针对3类驾驶风格和特征变量,采用不同长短期记忆神经(LSTM)网络训练,进一步强化LSTM参数特征,得到基于不同驾驶特征的轨迹预测模型和算法;通过实验室仿真列车运行数据对所提出的方法进行验证。验证结果表明:与传统单一算法相比,所提出的高速列车追踪运行轨迹预测方法具有显著优势,预测时长15 s内的平均相对误差MRE不超过3.5%,比标准LSTM和卡尔曼滤波器(KF)算法的MRE分别降低了32.9%和43.3%,且最短追踪时间间隔缩小0.3 min以上,该方法在一定程度上为满足西部铁路的特殊运能需求提供理论依据。In response to the significant differences in driving characteristics and trajectories of high⁃speed railway trains in different application scenarios in the western region in China,this paper established a high⁃speed railway train track⁃ing trajectory prediction model based on driving feature classification.Firstly,with the driving characteristics of high⁃speed train defined,three types of common and different driving styles of high⁃speed trains were identified based on gaussian mixture model(GMM)model combined with the characteristic data of tracking trains,such as speed,accelera⁃tion,jerk,time and space interval,while the importance characteristics by maximal information coefficient(MIC)of driving styles were analyzed.Secondly,different long short term memory(LSTM)neural networks were used to train three types of driving styles and feature variables,further enhancing the LSTM parameter features,and obtaining trajec⁃tory prediction models and algorithms based on different driving characteristics.Finally,the proposed method was verified and analyzed by using laboratory simulation of train operation data.The results show that compared with tradi⁃tional single algorithms,the proposed personalized trajectory prediction method shows significant advantages,its MRE does not exceed 3.5%within 15 s,with the MRE lower 32.9%and 43.3%than that of the standard LSTM and Kalman filter algorithms,respectively,and with a reduction of the shortest tracking time interval by more than 0.3 min,which can provide a theoretical basis for the special transportation capacity requirements of the western railways in China.

关 键 词：高速列车 轨迹预测 驾驶特征 长短期记忆神经网络 高斯混合模型 

分 类 号：U283.1[交通运输工程—交通信息工程及控制]