HAM-MSTF: A hybrid attention-based model for multi-task spatio-temporal forecasting

Area of Science:

• Data Science
• Artificial Intelligence
• Machine Learning

Background:

• Spatiotemporal forecasting is vital for decision-making across industries.
• Existing single-task models struggle with the complexity of real-world multi-task systems.
• Interrelationships among multiple forecasting tasks are often overlooked.

Purpose of the Study:

• To propose a novel hybrid attention-based model for multi-task spatiotemporal forecasting (HAM-MSTF).
• To address the limitations of single-task models by jointly predicting multiple variables.
• To enhance prediction performance by capturing inter-task relationships.

Main Methods:

• Utilized dual attention mechanisms for enhanced spatiotemporal feature extraction.
• Employed a graph attention mechanism within a multi-gate mixture-of-experts framework for spatial feature enrichment.
• Integrated global and local temporal dynamics with task-specific adaptation for independent spatiotemporal feature fusion.

Main Results:

• The HAM-MSTF model demonstrated excellent performance across multiple tasks on synthetic and benchmark datasets (PEMS04, PEMS08).
• Achieved superior multi-task coordination capabilities compared to existing spatiotemporal forecasting methods.
• Effectively captured complex spatial and temporal patterns while accounting for inter-task differences.

Conclusions:

• The proposed HAM-MSTF model offers a robust solution for multi-task spatiotemporal forecasting.
• Hybrid attention mechanisms and multi-task learning significantly improve prediction accuracy and coordination.
• This approach enhances operational efficiency and decision-making in various spatiotemporal applications.