Short term demand forecasting of electric vehicle charging stations using context aware temporal transformer model

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Rising electric vehicle (EV) adoption strains power grids, necessitating accurate demand forecasting.
• Existing research often overlooks location-specific charging behavior patterns.
• Understanding diverse charging patterns across different city locations is essential for grid management.

Purpose of the Study:

• To develop and evaluate a novel Context-Aware Temporal Transformer (CAT-Former) model for short-term EV charging demand forecasting.
• To incorporate temporal and contextual features to capture location-specific charging behaviors.
• To assess the model's performance against established baseline and transformer-based models.

Main Methods:

• Utilized public EV charging data from three high-traffic locations in Boulder City, Colorado.
• Developed the CAT-Former model, integrating temporal and contextual features for enhanced prediction.
• Compared CAT-Former's performance against LSTM, BiLSTM, CNN-LSTM, CNN-BiLSTM, and Transformer models using MSE and MAE metrics.

Main Results:

• The CAT-Former model significantly outperformed baseline and other transformer models in forecasting EV charging demand.
• Achieved the lowest Mean Square Error (MSE) and Mean Absolute Error (MAE) for one-hour and one-day ahead predictions.
• Demonstrated the effectiveness of temporal and contextual features in diverse charging environments.

Conclusions:

• The CAT-Former model provides a robust solution for short-term EV charging demand forecasting.
• Accurate forecasting using location-specific data can mitigate power supply challenges caused by EV charging.
• The study highlights the importance of context-aware models for smart grid management.