Optimizing Extended Metro Timetables During Year-End Events Using Graph Neural Networks: A Study on Passenger Flow Prediction and Timetable Optimization

Metropolitan areas with high public transportation ridership often face significant congestion, particularly during major national events. This study examines the increased passenger flow during year-end events in South Korea and proposes an optimized metro timetable to mitigate this issue. To achieve this, we employ the Spectral Temporal Graph Neural Network (StemGNN) to predict passenger flow at hub stations near event venues in Seoul and Busan. The results demonstrate that StemGNN outperforms traditional statistical models and stacked Long Short-Term Memory (LSTM) networks, confirming its effectiveness for this task. Using these predictions, we design an objective function that balances metro congestion and passenger convenience. The timetables of hub stations are optimized by minimizing the objective function value through Monte Carlo Simulation (MCS). The optimized timetable addresses issues such as excessive time intervals in high-demand sections and overly frequent intervals in low-demand sections, effectively reducing congestion and improving passenger convenience. On average, the optimized timetable achieves approximately a 36.843% reduction in congestion and a 69.782% reduction in passenger discomfort during peak congestion periods compared to the original timetable. Despite the limitation of data collection being confined to specific year-end events, this study provides valuable insights into improving metro operations. It contributes to enhancing the efficiency of public transportation and passenger satisfaction during peak periods.