铁道科学与工程学报2025，Vol.22Issue(5)：2027-2039,13.DOI:10.19713/j.cnki.43-1423/u.T20241099
拥堵条件下城市轨道交通"乘客-列车"分配方法
Passenger-to-train assignment method for urban rail transit under congestion
摘要
Abstract
Achieving fine-grained passenger assignment in urban rail transit under congested conditions is of great significance for improving congestion management of passenger flow and assisting passengers in their travel decisions.To this end,this paper focused on the line level and proposes a passenger-to-train assignment method integrating deep learning and knowledge reasoning,considering the uncertainty of the trains taken by passengers due to congestion delays.First,a constrained dual-channel fully connected neural network model was constructed using feature learning to estimate the entering(i.e.,access and waiting)and leaving(i.e.,egress)process times of passengers.Second,according to the time estimates,a train inference model was established based on the similarity measurement,and the discrete assignment probabilities were computed for the entire set of feasible trains,to assign passengers to their most likely trains that can be taken.Finally,to overcome the data heterogeneity bias,several bias correction strategies were formulated by integrating the above models and combined with domain knowledge to improve the assignment accuracy.The automatic fare collection and train timetable data of Chongqing Rail Transit Line 3 were used for a real-world case study.The results show that without relying on any independent assumptions or additional data,the proposed method can achieve a rational assignment for at least 87.12%of targeted passengers,which is about 12.50 percentage points higher than the baseline model without using the bias correction strategy,implying good applicability and assignment performance.Compared with the benchmark model constructed based on existing state-of-the-art methods,the consistency of assignment results is 90.12%,which suggests that the results are highly credible.Meanwhile,the computational efficiency of the proposed method after deployment is significantly higher than that of the benchmark model,which can better cope with large-scale practical problems.In addition,the proposed method can synchronize the estimation of delay characteristics(e.g.,the distribution of the number of times left behind,the probability of train boarding,etc.)of passenger trips,so as to better satisfy the decision-making or travel needs of operation managers and passengers,and help to deepen the perception and understanding of the congestion dynamics of the passenger flow.Overall,this study can enriches the methodology development for the refined passenger flow assignment in urban rail transit and is able to provide important support for the passenger-to-train assignment at the network level.关键词
城市轨道交通/乘客-列车分配/深度学习/偏差纠正策略/拥堵延误特征Key words
urban rail transit/passenger-to-train assignment/deep learning/bias correction strategy/congestion delay characteristics分类
交通工程引用本文复制引用
闫旭,彭其渊,张永祥,刘晓薇,谷丽婷,冯涛..拥堵条件下城市轨道交通"乘客-列车"分配方法[J].铁道科学与工程学报,2025,22(5):2027-2039,13.基金项目
国家重点研发计划项目(2022YFB4300502) （2022YFB4300502）
国家自然科学基金资助项目(72201218) （72201218）
四川省科技计划资助项目(2023NSFSC0901) （2023NSFSC0901）
