Novel Hybrid Spatiotemporal Convolution Neural Network Model for Short-Term Passenger Flow Prediction in a Large-Scale Metro System

Abstract

Accurate and reliable prediction of subway passenger flow is a particularly challenging application of spatiotemporal forecasting, due to the time-varying travel patterns and the complex spatial dependencies on subway networks. To address these challenges, this study proposes a novel spatiotemporal graph convolutional bidirectional long short-term memory neural network model combined with an attention mechanism (At-STGCN-BiLSTM) to better predict short-term passenger flow for all stations in a large-scale metro system. The STGCN-BiLSTM aims to capture the attributes of spatiotemporal characteristics of subway stations, and the attention mechanism helps account for the correlation between historical data and current moment inbound passenger flow. The performance of the short-time passenger flow forecast model is analyzed by different time intervals. Experimental results show that the proposed model outperforms baseline models on Wuhan, China, subway data. The value of root-mean square error (RMSE) and mean absolute error (MAE) decreased by 7.33% and 9.38%, respectively, compared with the baseline models at the 15-min interval. The attention mechanism in the proposed model can effectively improve the prediction capability of peak and nonperiodic passenger flow variations. The research not only is of great help to the passenger flow organization and emergency management of the subway, but also plays a vital role in the work of rail transit regulation, rail transit alarm release, and service efficiency improvement. This research introduces a novel approach for predicting subway passenger flow, offering valuable insights for both transport authorities and commuters. By considering the spatiotemporal dynamics of passenger movement and incorporating an attention mechanism, the proposed model enhances short-term flow predictions. In practical terms, this means more accurate estimations of passenger numbers, travel times, and congestion levels for subway stations. The model’s adaptability to varying scenarios, including peak hours and unexpected disruptions, ensures reliable real-time predictions. For transit operators, this model aids in optimizing resource allocation, enhancing service efficiency, and facilitating emergency management. Commuters benefit from improved trip planning and a smoother travel experience. Beyond the subway context, the methodology’s fusion of advanced techniques showcases its potential to inform broader transportation systems and urban planning.