Predicting subway passenger flows under different traffic conditions.

Area of Science:

• Urban transportation systems analysis
• Data science and predictive modeling
• Smart city infrastructure management

Background:

• Effective operation, management, and reliability of urban rail transit systems depend on accurate passenger flow prediction.
• Large-scale smartcard data offers a rich resource for understanding and forecasting subway passenger dynamics.
• Differentiating between ordinary and anomalous traffic conditions is crucial for robust prediction.

Purpose of the Study:

• To predict dynamic passenger flows within a major urban subway network using smartcard data.
• To evaluate the performance of four classical predictive models under varying traffic conditions (ordinary vs. anomalous).
• To determine the optimal prediction horizon for each model and identify factors influencing prediction accuracy.

Main Methods:

• Utilized large-scale smartcard data from Shenzhen's subway system.
• Applied four predictive models: historical average, multilayer perceptron neural network, support vector regression, and gradient boosted regression trees.
• Employed the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to identify ordinary and anomalous traffic conditions at subway stations.

Main Results:

• Prediction accuracy varied significantly across models and traffic conditions.
• Model performance was assessed under both ordinary and anomalous traffic scenarios.
• The study investigated the lead time for accurate passenger flow predictions by each model.

Conclusions:

• The selection of appropriate predictive models is critical for enhancing passenger flow prediction accuracy.
• Inherent patterns within passenger flow data play a more significant role in prediction accuracy than traffic conditions alone.
• Understanding these patterns and model-specific performance is key to optimizing urban rail transit management.