A novel hybrid deterministic temporal fusion transformer and dynamic recurrent neural network architecture for multi-horizon multi-variate electric vehicle energy consumption forecasting: mitigation pathway

Integration of distributed renewable energy sources and penetration of Electric Vehicles (EVs) becomes inevitable to mitigate sharp global warming. Due to the inherent intermittent nature of EVs, accurate prediction of energy consumption is crucial in attaining optimal energy management. The least modelled real-time multi-step EV energy consumption prediction problem remains challenging due to the complex driving patterns, external factors, and vast diversity in EV charging pattern. To solve complex multi-horizon multi-variate problem, a unique framework integrating Deterministic Temporal Fusion Transformer (DTFT) model, Recurrent Neural Network (RNN) and black widow metaheuristic optimization algorithm is proposed in the research article. To provide insight to the environmental benefits of optimal energy usage, multi-horizon Greenhouse Gas (GHG) savings were also predicted. Performance evaluation metrics including R-square, concordance correlation coefficient, Mean Absolute Error (MAE), Normalize Root Mean Square Error (NRMSE) and other key metrics validates the robustness and accuracy of the proposed method in multi-horizon multi-variate prediction.