Analysis And Application Of Train Choice Behavior For High-Speed Railway Passenger Based On Deep Learning

With the continuous development of high-speed railway(HSR)network,Chinese HSR is gradually shifting stage from ‘construction’ to ‘operation’ at the same time passengers’ demand for the quality of transport products and personalized services are gradually prominent.Besides,Chinese HSR network is extensive in scale,the structure of passenger flow is complex,and the travel demand of different passengers are significantly different,meanwhile the train also present different characteristics in terms of origin and destination,travel speed,stops,and service level etc,which provids passengers with more choices.Therefore,a more accurate grasp of the characteristics of passengers’ travel choices behavior for trains,then optimize product elements purposeful,which has important theoretical and practical significance for improving the quality of HSR transport products.To solve this problem,this paper applies HSR ticket data to study passengers’ travel choice behavior for trains based on big data technology and Deep Learning,providing more realistic loading rules for the passenger flow assignment decision in the design of train operation plan.Firstly,the paper analyzes factors affect HSR passengers’ travel choice behavior for trains.Travelers have different choice behavior at different time,which determines the model’s input data in the following study.Through the analysis of the influence of passengers-related attributes on travel choices,disposable income and travel distance are chosen as the elements of HSR OD clustering.The influence of train-service related factors on travel choices are evaluated,and further taken as the follow-up Deep Learnning model input characteristics.Considering the considerable HSR OD,the complex passenger flow structure and the significant differences in passenger travel demand,the multi-step central clustering is carried out to obtain several OD sets with relatively similar travel choice behavior by the factors of passengers-related attributes,OD passenger flow and the frequency of OD,then study passengers’ s travel choice behavior and form the passenger flow assignment loading rules by OD category to improve the accuracy of the model.On the basis of HSR OD clustering,this paper constructs HSR passengers train choice behavior model with ticket data based on the Deep Learning framework,and study the key parameters’ s selection such as network layers and hidden nodes,and analyzes the optimization method of model.Considering the subjective contingency of passengers’ choice,on the basis of the average accuracy,the paper proposes the overall satisfaction rate,which can evaluate the model more comprehensivly.Finally,taking the summer passenger ticket data in 2016 and 2017 as example,the HSR passenger travel train choice behavior Deep Learnning model is verified by actual cases,then analyze the influence of train-service related factors and proposes new passenger flow assignment strategies based on the optimized model,which can further be applied for the optimization of train operation plan.Simultaneously the passenger flow assignment method based on the prediction results of Deep Learning model is compared with the traditional Logit model.Results show that the method applied in this paper is more consistent with the actual situation.