Multi-scale fusion transformer for EV charging station load prediction

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Electric vehicle (EV) charging load prediction is essential for power grid stability and infrastructure optimization.
• Challenges include the unpredictable nature of charging and external influences like weather and traffic.

Purpose of the Study:

• To develop a novel Transformer-based architecture for accurate EV charging station load prediction.
• To enhance prediction by integrating multi-scale temporal modeling and external feature analysis.

Main Methods:

• Proposed the Multi-scale Fusion Transformer (MFT) model.
• Integrated a Multi-scale Modeling Mechanism (3M) for temporal dependencies.
• Employed a Feature-correlation Analysis Module (FAM) for external factors (weather, traffic).
• Utilized a Multi-variable Fusion Module (MFM) with cross-attention for dynamic feature fusion.

Main Results:

• MFT significantly outperformed baseline models in both short-term and long-term load forecasting.
• Achieved up to 25.59% average performance improvement, demonstrating superior accuracy and stability.
• Validated using real-world data from Norway.

Conclusions:

• The MFT model effectively captures complex, multi-scale, and externally influenced load patterns.
• Offers a robust solution for intelligent grid scheduling and energy management in the era of widespread EV adoption.