Short-Term Passenger Flow Prediction in Urban Rail Transit Based on Points of Interest

In the rapidly evolving landscape of smart transportation, the passenger volume in urban rail transit consistently demonstrates an upward trajectory. In this context, precise and scientifically grounded short-term passenger flow prediction methods are essential for optimizing operational scheduling and ensuring safety in urban rail transit. Consequently, this paper introduces Temporal Graph Attention Long Short-Term Memory (TGALSTM), a spatiotemporal integrated prediction network model that incorporates the surrounding environment of the station. Initially, the paper enhanced the Temporal Convolutional Network (TCN) model to capture temporal features more accurately. Subsequently, the paper utilizes the Graph Attention Network (GAT) network module specifically to extract the topological structure and surrounding environmental features of the station. Lastly, the prediction task is accomplished by weighted fusion of various features, inputting them into the Attention Long Short-Term Memory (LSTM) network. Experiments were conducted on two authentic datasets, revealing that the TGALSTM model outperforms the baseline model in both single-step and double-step predictions, showcasing the model's exceptional performance and robustness. This research offers a robust method and support to enhance the operational efficiency and passenger flow management of urban rail transit systems.