基于多分支ResCovLSTM的城市轨道交通短时客流预测模型  

Short-term passenger flow prediction model for urban rail transit based on multi branch ResCovLSTM

作　　者：刘燕[1] 李恒如 谷卫 LIU Yan;LI Hengru;GU Wei(Rail Transit College of Changzhou Railway Vocational and Technical College,Changzhou Jiangsu 213000,China;Changzhou Metro Group Co.,Ltd.,Changzhou Jiangsu 213000,China)

机构地区：[1]常州铁道高等职业技术学校轨道交通学院,江苏常州213000 [2]常州地铁集团有限公司,江苏常州213000

出　　处：《现代城市轨道交通》2025年第2期130-139,共10页Modern Urban Transit

摘　　要：随着城市化进程的加速,城市轨道交通客流预测对于提高运营效率和服务质量愈发重要。然而,现有模型在处理大规模、多维度数据时面临预测精度不足和计算复杂度高的挑战。为解决该问题,文章提出一种基于多分支ResCovLSTM的深度学习模型,创新性地设计4个独立分支,分别处理天气与空气质量、流入量、流出量以及网络拓扑结构等关键因素。通过融合残差网络、CovLSTM和卷积注意力等模块,有效提升预测精度和模型泛化能力。实验结果表明,该模型在单步和多步预测中均表现出色,显著降低预测误差。以WMAPE为例,模型在单步预测中的WMAPE仅为8.625 1%,相比次优模型降低0.16%,证明模型的有效性和优越性。With the acceleration of urbanization,passenger flow prediction of urban rail transit is crucial for improving operational efficiency and service quality.However,existing models face challenges of insufficient prediction accuracy and high computational complexity when dealing with large-scale,multidimensional data.To address this issue,this article proposes a deep learning model based on multi branch ResCovLSTM,innovatively designs four independent branches to handle key factors such as weather conditions and air quality,inflow,outflow,as well as network topology.By combining residual networks,CovLSTM and convolutional attention modules,the model effectively improves prediction accuracy and generalization ability.The test results show that the model proposed in this article performs well in both single step and multi-step prediction,significantly reducing prediction errors.Taking WMAPE as an example,the WMAPE of our model in single step prediction is only 8.6251%,which is 0.16%lower than the suboptimal model,proving the effectiveness and superiority of the model.

关 键 词：城市轨道交通 短时客流预测 多步预测 深度学习 ResCovLSTM 

分 类 号：U293.13[交通运输工程—交通运输规划与管理]