Individual response prediction and personalized guidance strategy optimization in urban rail transit networks

Advanced travel information systems play a crucial role in alleviating network-wide congestion in urban rail transit. However, existing studies overlook the heterogeneity of passengers’ compliance with information and their personalized requirements, leading to inefficiencies in guidance. To address these limitations, this paper explores the personalized guidance problem. An integrated framework is proposed to strategically provide passengers with differential route suggestions, ultimately minimizing systematic cost. The framework includes two modules, the first of which predicts individual route decisions under information. Passenger preferences are incorporated into the gradient boosting decision tree model to capture the heterogeneity of compliance with information. Additionally, this module integrates automated fare collection data with stated preference data, thereby avoiding the large-scale and costly data collection. The second module formulates and solves the personalized guidance problem. The problem is modeled as a Markov decision process encompassing an extensive solution space. Moreover, the deep deterministic policy gradient approach is utilized to overcome the dynamicity and dimensional disaster of the problem. A case study of the Beijing Subway is provided to highlight the effectiveness of the proposed framework. The findings show that the guidance strategy significantly decreases the network-wide generalized travel cost by 18.7%, with considerable benefits in overcrowded regions by guiding passengers toward less crowded areas. Moreover, the proposed framework accurately predicts individual behavior responses in route choice, reducing the mean squared error by at least 18.5 %. This study offers valuable information for subway managers to effectively organize passenger flow and improve the quality of passenger travel.