A graph spatial-temporal model for predicting population density of key areas.

Area of Science:

• Urban planning
• Data science
• Epidemiology

Background:

• Accurate population density prediction is vital for managing urban areas.
• Existing methods often rely on regional images or clustering, with limited exploration of spatial-temporal models.
• Predicting population density aids in controlling disease spread (e.g., Covid-19) and understanding travel patterns.

Purpose of the Study:

• To propose a novel spatial-temporal model for predicting urban population density without relying on regional images.
• To introduce the Word Embedded Spatial-temporal Graph Convolutional Network (WE-STGCN) for this purpose.
• To evaluate the model's performance against established methods.

Main Methods:

• Representing 997 key urban areas and their connections as a graph structure.
• Developing the WE-STGCN model, incorporating Spatial Convolution Layer, Temporal Convolution Layer, and Feature Component.
• Utilizing a dataset from the DataFountain platform for evaluation.

Main Results:

• The WE-STGCN model demonstrated a significant improvement over baseline models.
• Average performance improvement reached 53.97% compared to existing methods.
• The model proved effective in commendably predicting population density in key urban areas.

Conclusions:

• The proposed WE-STGCN model offers a powerful new approach for spatial-temporal population density prediction.
• This method enhances the capability to manage urban dynamics, disease spread, and transportation needs.
• The model's effectiveness is validated through comparative experiments, highlighting its potential for practical applications.