Short-Term Passenger Flow Prediction for Urban Rail Based on Improved EEMD-Ensemble Learning

Urban rail transit passenger flow prediction encounters two primary challenges: the inability to effectively capture the variations in passenger flow patterns and the limited accuracy of prediction models. Therefore, this paper proposes three improved Ensemble Empirical Mode Decomposition (EEMD) methods to accurately capture the characteristics of passenger flow sequence data: (1) Grouped Reconstruction of IMF (Intrinsic Mode Function) Components in EEMD; (2) High-Frequency Disturbance Removal EEMD; (3) Hybrid EEMD. Simultaneously, a short-term passenger flow combination prediction model for urban rail transit stations is established. In order to enhance the prediction accuracy and robustness of the model, the base models utilized include XGBoost, LightGBM, and LSTM. Additionally, the Stacking ensemble method is employed to fuse the models together. Finally, a short-term passenger flow prediction model for urban rail transit based on improved EEMD and ensemble learning is constructed. Taking the entry passenger flow of Beijing West Station in the Beijing Subway as an example, the effectiveness of the model in predicting short-term passenger flow in urban rail transit is validated.