Towards Adaptive Energy Intelligence using Deep Learning based Battery Management and Charging Duration Optimization for Electric Vehicles

Abstract

Electric vehicles (EVs) are central to the shift toward a greener, more sustainable global economy, yet optimal charging-duration prediction and efficient battery management remain persistent challenges. Estimating a battery's remaining life span helps users gauge driving range, while accurately forecasting charging energy needs benefits manufacturers and owners alike improving usage rates, cutting costs, and easing the load on electric grid stations. Station-level energy consumption is also shaped by factors such as EV adoption trends, plug-in timing, random usage patterns, and differences between holidays and working days. To address overcharging, energy wastage, and the broader challenge of managing dispatch stations through data-driven analysis, this paper proposes a Deep Stacking Ensemble for Battery Management and Optimal Charging Duration in Electric Vehicles (DSEBM-OCDEV). The method aims to build an intelligent battery management framework that predicts optimal charging duration using a hybrid ensemble learning approach. Raw battery data are first pre-processed through outlier removal, categorical encoding, and feature scaling for consistency and stability. Correlation-based feature selection then identifies the most informative battery attributes while reducing redundancy and computational overhead. For charging-duration classification, a stacking ensemble combines a variational graph autoencoder, a spiking neural network, and a double deep Q-network, with their outputs merged through a meta-learner optimized using AdamW to improve convergence and generalization. Extensive simulation results confirm that DSEBM-OCDEV consistently outperforms existing methods.