An urban crowd flow model integrating geographic characteristics.

Area of Science:

• Urban planning and management
• Geographic information systems
• Data science and artificial intelligence

Background:

• Accurate urban crowd flow prediction is crucial for public safety and traffic management.
• Existing models often overlook geographic characteristics, limiting their spatial dependency modeling.
• There is a need for advanced methods that incorporate geographical features for better crowd flow forecasting.

Purpose of the Study:

• To propose a novel urban crowd flow prediction model (FPM-geo) that integrates geographic characteristics.
• To enhance the modeling of spatial dependency between urban regions.
• To improve the accuracy of short-term crowd flow predictions.

Main Methods:

• A residual multigraph convolution network is used to fuse proximity, functional similarity, and road network connectivity.
• A long short-term memory network integrates local crowd flow dynamics and spatial dependencies.
• A 4-day mobile phone dataset was utilized for model validation.

Main Results:

• The proposed FPM-geo model demonstrated a 15.37% reduction in root mean square error compared to traditional models at a 15-minute prediction interval.
• Prediction error correlates positively with local crowd flow volume.
• Prediction error peaks during morning and evening rush hours and is lowest at night.

Conclusions:

• Integrating geographic characteristics significantly enhances urban crowd flow prediction accuracy.
• The FPM-geo model effectively captures both spatial dependencies and temporal dynamics.
• The findings provide valuable insights for optimizing urban management strategies through improved crowd flow forecasting.