Bidirectional Long Short-Term Memory Model for Metro Passenger flow Prediction

To maximize metro operations, accurate short-term passenger flow projections are essential. The paper utilizes Bidirectional Long Short-Term Memory (BiLSTM) to forecast passenger flow based on data from January 16 to January 25, 2019. By contrasting BiLSTM with conventional models of Long Short-Term Memory (LSTM) and Recurrent Neural Network (RNN), the research highlights BiLSTM’s superior ability to capture temporal dependencies from both past and future data. The analysis reveals distinct patterns for weekdays and weekends, with double peaks during commute hours on weekdays and a continuous peak in the afternoon on weekends. The results indicate that BiLSTM outstandingly enhances the prediction's accuracy, with less Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) in contrast to RNN and LSTM. This enhanced predictive capability supports more effective scheduling and flow management in metro systems, ensuring better service and operational efficiency. The study underscores the practical benefits of BiLSTM in handling complex, dynamic passenger flow data.