Optimizing Electric Vehicle Demand Forecasting in Electrical Transportation System Using a Hybrid GRU- Self-Attention Framework

The rapid rise in electric vehicle (EV) adoption necessitates efficient forecasting of EVs charging load to ensure stable grid operations and optimized energy source allocation. Existing approaches such as statistical and standalone deep learning models, often fall short in capturing both short-term and long-term dependencies. However, hybrid models demand significant computational resources, necessitating lightweight, forecasting frameworks suitable for edge devices in smart grids. This study introduces a hybrid GRU-Self-Attention model that utilizes Gated Recurrent Units (GRUs) for sequential learning and the self-attention mechanism for long-range dependency capture. The model uses the open-source dataset of EVs charging sessions, and demonstrates superior performance in EVs load demand forecasting. The proposed hybrid model achieves a significant reduction in Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) compared to Long Short-Term Memory (LSTM) and BiLSTM models. This work provides a scalable and efficient framework for EV load prediction and contributes to hybrid architecture for EVs demand forecasting.