Graph convolutional network approach applied to predict hourly bike-sharing demands considering spatial, temporal,...

Area of Science:

• Urban planning
• Transportation engineering
• Data science

Background:

• Public bike-sharing systems face challenges with supply-demand imbalances.
• Accurate demand prediction is crucial for system stability and efficiency.
• Existing models often fail to simultaneously consider spatial and temporal factors in bike demand.

Purpose of the Study:

• To develop an advanced prediction framework for public bike-sharing demand.
• To improve prediction accuracy by integrating spatial and temporal features.
• To account for external factors influencing bike demand.

Main Methods:

• Proposed a novel prediction framework utilizing graph convolutional networks (GCNs).
• Incorporated spatial dependencies between bike-sharing stations.
• Modeled diverse temporal patterns and integrated global variables like weather and day type.

Main Results:

• The proposed GCN-based framework demonstrated superior performance compared to baseline models.
• Simultaneous consideration of spatial and temporal properties significantly enhanced prediction accuracy.
• Integration of global variables further refined the model's predictive capabilities.

Conclusions:

• The developed framework effectively addresses the supply-demand imbalance in bike-sharing systems.
• Graph convolutional networks offer a powerful approach for spatiotemporal bike demand forecasting.
• This research provides a robust solution for optimizing public bike-sharing operations.