Share
Export Citation
Interpretable Ensemble Voting Systems for State of Charge Prediction in Electric Vehicle Batteries
Zhipeng F.
Cybernetics and Systems
Q3Abstract
The accurate prediction of the state of charge (SoC) in electric vehicle (EV) batteries is a critical challenge due to lithium-ion batteries’ complex, nonlinear behavior under diverse operating conditions. While capable of achieving high prediction accuracy, existing methods often lack interpretability, rendering them impractical for real-world applications. We propose a novel framework, interpretable ensemble voting systems for SoC Prediction, to address this dual challenge of performance and transparency. The framework integrates an Ensemble Voting Regressor (EVR) for SoC prediction, leveraging eight ensemble algorithms to identify the two top-performing models. The EVR subsequently integrates these two models into a unified system to enhance prediction accuracy, achieving R-squared scores of 0.97 and 0.99. To overcome the interpretability gap, an Explainable Voting System (EVS) is introduced, incorporating SHapley Additive exPlanation (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). Through global interpretability analysis, the EVS identified critical parameters influencing SoC—such as Battery Voltage, Battery Current, Motor Torque, and Throttle—while pinpointing inflection points that accelerate SoC transitions. This study lays the groundwork for developing interpretable and high-performance SoC prediction methods. By combining predictive accuracy with actionable insights, the proposed framework offers potential implications for optimizing EV performance and supporting informed decision-making in the field.