Enhanced Energy Consumption Prediction for Electric Vehicles Using Attention-Augmented GRU and Ensemble Learning

The rapid proliferation of electric vehicles (EVs) necessitates accurate prediction of energy consumption to alleviate range anxiety and optimize battery management. This paper proposes an advanced framework for predicting EV energy consumption ($\mathbf{k W h} / \mathbf{1 0 0 k m}$) under diverse driving conditions. Leveraging a Kaggle dataset, we introduce feature engineering, such as speed change rate, and employ machine learning models including Random Forest, eXtreme Gradient Boosting (XGBoost), Gated Recurrent Unit (GRU), attention-augmented GRU, and an ensemble method. The methodology incorporates time-series sequencing with a window of 10, hyperparameter tuning via randomized search, and early stopping for efficient training. Experimental results demonstrate that the ensemble model achieves the lowest Mean Squared Error (MSE) of 0.3215 and highest Coefficient of Determination ($\mathbf{R}^{2}$) of $\mathbf{0. 9 2 5 6}$, outperforming baselines by up to $20 \%$. Ablation studies confirm the attention mechanism’s contribution, reducing MSE by $8.22 \%$. This work enhances EV energy efficiency, supporting sustainable transportation. Future extensions could integrate real-time sensor data for deployment in navigation systems.