MTMF-Grid: A multi-task multi-modal fusion model for operational forecasting and decision support in power grids

Area of Science:

• Electrical Engineering
• Data Science
• Investment Strategy

Background:

• Power grid investments involve complex, multi-objective decisions with heterogeneous data.
• Traditional single-task or single-modal methods are insufficient for precise strategic investment decision-making.

Purpose of the Study:

• To propose a novel multi-task multi-modal fusion model (MTMF-Grid) for operational forecasting and decision support in strategic emerging power grid investments.
• To leverage operational data proxies for informed investment decisions, bypassing direct financial return prediction.

Main Methods:

• Developed a modular MTMF-Grid architecture.
• Implemented a task-adaptive Transformer for balanced feature expression.
• Utilized a Cross-Fusion Gating Mechanism (CFGM) for robust multi-modal fusion, handling missing data.
• Employed a loss variance-based mechanism for dynamic task weight adjustment and gradient conflict mitigation.

Main Results:

• MTMF-Grid demonstrated superior performance on SEWA and OPSD datasets compared to baseline models.
• Achieved 3.06% Mean Absolute Percentage Error (MAPE) for hourly electricity price prediction.
• Reached 0.915 accuracy for load fluctuation risk classification.

Conclusions:

• The proposed MTMF-Grid offers a comprehensive framework for strategic power grid investment decision-making.
• Effective integration of operational forecasting and multi-modal data enhances investment support.
• The model's robustness and adaptability are key contributions to the field.