Spatiotemporal Graph Convolution Multifusion Network for Urban Vehicle Emission Prediction.

Area of Science:

• Environmental Science
• Computer Science
• Urban Planning

Background:

• Urban vehicle emissions pose significant challenges for pollution control and traffic management.
• Existing deep learning methods struggle with the complex spatiotemporal dynamics and graph-structured nature of road networks.

Purpose of the Study:

• To develop an advanced model for accurate urban vehicle emission prediction.
• To address the limitations of current methods by incorporating graph structures and external environmental factors.

Main Methods:

• Proposed a spatiotemporal graph convolution multifusion network (ST-MFGCN).
• Employed graph convolution to model spatial dependencies and temporal convolution for temporal correlations.
• Integrated multisource external factors (global and individual features).
• Utilized a multifusion strategy to merge spatiotemporal patterns and external data.

Main Results:

• The ST-MFGCN model effectively captures spatiotemporal dependencies in vehicle emissions.
• The model successfully learns the impact of complex external environmental factors.
• Evaluated on real-world vehicle emission data, demonstrating effective prediction capabilities.

Conclusions:

• The ST-MFGCN provides a robust framework for urban vehicle emission prediction.
• This approach enhances the understanding and management of vehicle pollution in urban environments.