FF-STGCN: A usage pattern similarity based dual-network for bike-sharing demand prediction.

Area of Science:

• Transportation Science
• Data Science
• Urban Planning

Background:

• Bike-sharing systems face challenges in demand prediction due to complex spatio-temporal user behavior.
• Unbalanced bike distribution arises from arbitrary user choices, impacting system efficiency.
• Accurate prediction is vital for bike allocation, rebalancing, and strategic station planning.

Purpose of the Study:

• To propose a novel dual-network model, FF-STGCN, for accurate bike-sharing demand prediction.
• To effectively integrate inter-station flow and similar usage pattern features into the prediction model.
• To address limitations in multi-scale spatio-temporal accuracy for improved bike-sharing management.

Main Methods:

• Developed a multi-scale spatio-temporal feature fusion module to enhance accuracy.
• Constructed a bike usage pattern similarity learning module to capture station correlations.
• Employed a dual-network structure integrating flow and pattern features for final demand prediction.

Main Results:

• The FF-STGCN model demonstrated significant effectiveness on the Citi Bike dataset.
• Ablation experiments confirmed the crucial contribution of each module within the proposed model.
• The model successfully integrated diverse features for more accurate bike-sharing demand forecasting.

Conclusions:

• The proposed FF-STGCN model offers an effective solution for bike-sharing demand prediction.
• Integrating inter-station flow and usage pattern similarities significantly improves prediction accuracy.
• This approach provides a valuable tool for optimizing bike-sharing system operations and planning.