Charging stations demand forecasting using LSTM based hybrid transformer model.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate electric vehicle (EV) energy demand forecasting is essential for power system stability and reliable charging station operation.
• Medium- and long-term predictions are crucial for analyzing charging demand patterns based on historical data.

Purpose of the Study:

• To propose and evaluate a novel LSTM-based encoder-decoder Transformer model for forecasting electric vehicle charging station (EVCS) demand.
• To compare the proposed model's performance against traditional LSTM and Transformer models using real-world datasets.

Main Methods:

• Developed a hybrid LSTM-Transformer model integrating an LSTM-based encoder-decoder architecture.
• Trained and tested the model on open datasets from ACN, specifically Caltech and JPL charging data.
• Evaluated performance for 30, 120, and 240-day ahead demand predictions using Mean Absolute Error (MAE) and Mean Squared Error (MSE).

Main Results:

• The LSTM-Transformer model demonstrated significant improvements over baseline models for both Caltech and JPL datasets.
• For Caltech data, MAE and MSE were reduced by up to 17.27% and 19.79% at the 30-day horizon, respectively.
• For JPL data, MAE and MSE reductions reached up to 24.91% and 23.17% at 30 days, with consistent improvements across longer horizons.

Conclusions:

• The proposed LSTM-Transformer model effectively enhances medium- and long-term EV charging demand forecasting accuracy.
• The hybrid model outperforms traditional deep learning approaches, offering more reliable predictions for power system management and charging infrastructure planning.