基于ATT-CNN-BiLSTM的虚拟编组列车时空轨迹预测  

Time-Position Trajectory Prediction of Trains in Virtual Coupling Based on ATT-CNN-BiLSTM

作　　者：柴铭[1,2] 刘皓元 苏浩翔 唐涛 刘宏杰[1,2] CHAI Ming;LIU Haoyuan;SU Haoxiang;TANG Tao;LIU Hongjie(State Key Laboratory of Advanced Rail Autonomous Operation,Beijing Jiaotong University,Beijing 100044,China;National Engineering Research Center of Rail Transportation Operation and Control System,Beijing Jiaotong University,Beijing 100044,China;School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]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 [2]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 [3]北京交通大学电子信息工程学院,北京100044

出　　处：《铁道学报》2024年第6期80-89,共10页Journal of the China Railway Society

基　　金：国家自然科学基金(52372309,52372310);先进轨道交通自主运行全国重点实验室(北京交通大学)(RAO2023ZZ001);中国铁路总公司实验室基础研究项目(L2021G009)。

摘　　要：保障虚拟编组平稳追踪运行的关键问题是实现对列车运行状态的精准预测。针对列车运行过程多变的特点,提出基于融合注意力机制的卷积双向长短期记忆神经网络(ATT-CNN-BiLSTM)的时空轨迹预测方法。针对列车历史运行数据中非正常运行场景稀少产生的数据非均衡问题,利用卷积神经网络和双向长短期记忆网络提取列车运行数据维度之间的特征关联,并增加注意力机制提升泛化能力。同时引入运行时验证方法在线监控预测结果,降低由预测错误造成的行车风险。以成都地铁8号线真实数据为例进行实验,设计5种评价指标,通过基线模型与消融实验对所提ATT-CNN-BiLSTM进行评价,该模型对于异常场景的预测误差至少减小9.626%。In virtual coupling,predicting operation states of trains accurately is a central problem in ensuring the smooth tracking of trains.Considering the ever-changing characteristics of train operations,a spatio-temporal trajectory prediction method was proposed based on convolutional bidirectional long short-term memory neural network with attention mechanism(ATT-CNN-BiLSTM).To address the problem of imbalanced data caused by few abnormal train operation scenarios in historical train operation data,convolutional neural network and bi-directional long short-term memory network were used to extract feature correlations between dimensions of train operation data,with attention mechanism added to enhance generalization ability.Meanwhile,the runtime verification method was introduced to monitor the prediction results online to reduce the operational risks caused by prediction errors.Based on the data of Chengdu Metro Line 8 for experiment,the ATT-CNN-BiLSTM model proposed in this paper was evaluated by baseline model and ablation experiment with 5 evaluation indexes.The results show that the prediction error of the model for abnormal scenes is reduced by at least 9.626%.

关 键 词：列车状态预测 虚拟编组 深度学习 注意力机制 双向长短期记忆神经网络 

分 类 号：U283.5[交通运输工程—交通信息工程及控制]