PREDICTING ELECTRIC VEHICLE BATTERY STATE OF CHARGE WITH ENSEMBLE LEARNING FOR ENERGY EFFICIENCY OPTIMISATION

Abstract

The transition from internal combustion engine vehicles to electric vehicles (EVs) is central to decarbonizing road transport, but limited driving range, battery degradation and the variability of real-world energy use continue to restrict adoption. Accurate, data-driven estimation of the energy state of the vehicle is a prerequisite for energy management strategies that can respond to changing driving and environmental conditions. This study models an intelligent energy efficiency system for EVs that organizes vehicle telemetry into a three-stage pipeline of data collection, data preprocessing and learning and evaluation, and uses the resulting predictions to drive energy optimization actions such as speed adjustment, battery optimization and power distribution. At the core of the system, three machine learning regressors, a Random Forest, an Extreme Gradient Boosting (XGBoost) model and a multilayer perceptron (MLP) neural network, were trained to predict the state of charge (SOC) of the high-voltage battery from cleaned driving, battery and environmental features. The models were evaluated on an 80:20 train–test split with fixed hyperparameters and further assessed by five-fold cross-validation. XGBoost achieved the best overall performance, with a root mean squared error (RMSE) of 0.7809, a coefficient of determination (R²) of 0.9724 and the lowest cross-validation RMSE of 0.8002, while the Random Forest attained the lowest mean absolute error (MAE) of 0.0834 and an R² of 0.9673. The MLP performed markedly worse, with an RMSE of 1.6011, an R² of 0.8838 and a cross-validation RMSE of 1.8326, indicating weaker generalization on the available data. The results show that tree-based ensemble learners provide accurate and stable SOC predictions that can serve as the predictive layer of software-based EV energy management, improving energy awareness without hardware modification.