Forecasting Subway Passenger Flow for Station-Level Service Supply

Area of Science:

• Artificial Intelligence
• Transportation Science
• Operations Research

Background:

• Accurate demand forecasting is essential for service supply chain management, particularly in predicting passenger flow at subway stations.
• Current forecasting models lack the ability to differentiate passenger flow based on varying station types, hindering effective service planning.

Purpose of the Study:

• To develop an advanced deep learning model for predicting subway passenger flow that accounts for diverse station functional types.
• To enhance the accuracy and robustness of subway passenger forecasting for improved service supply chain management.

Main Methods:

• Proposed a novel deep learning architecture named DeepSPF (Deep Learning for Subway Passenger Forecasting).
• Integrated sliding long short-term memory (LSTM) neural networks with one-dimensional convolution as a core component of the DeepSPF model.
• Utilized data from the Beijing subway system for experimental validation.

Main Results:

• DeepSPF demonstrated superior performance compared to baseline models across multiple time granularities (10, 15, and 30 minutes).
• The inclusion of station functional type information in DeepSPF significantly improved its robustness, especially during abnormal situations.
• The model accurately predicts future passenger flow at different station types.

Conclusions:

• DeepSPF offers a more accurate and robust solution for subway passenger flow forecasting by incorporating station-specific characteristics.
• The findings support the use of DeepSPF for optimizing service planning and resource allocation in subway systems.
• This research contributes to advancing demand forecasting techniques in the context of urban transportation.