Advanced Battery Management System for Electric Vehicles Using Artificial Intelligence Techniques

Abstract

Electric-vehicle battery packs operate under nonlinear, temperature-dependent and aging-sensitive conditions, making state estimation and safety control difficult for conventional rule-based battery management systems (BMSs). This methodology paper develops a reproducible artificial-intelligence-assisted BMS framework that jointly estimates state of charge (SOC) and state of health (SOH), detects electrical/thermal anomalies, and fuses the predictions with deterministic safety limits. The proposed study uses voltage, current, temperature and resistance-related indicators collected in a physics-informed EV battery simulation, with cycle-level health features used for slow-time SOH estimation. Random Forest, Gradient Boosting and multilayer perceptron (MLP) models are benchmarked and combined through weighted ensemble fusion. A separate Random Forest classifier performs anomaly detection using instantaneous variables and temporal derivatives. Group-wise train/test separation is used to prevent samples from the same simulated cycle appearing in both sets. In the executed benchmark, the ensemble achieved SOC MAE of 0.79% and SOH MAE of 0.53%, while the anomaly detector attained an F1-score of 0.995 on injected fault events. These results demonstrate the methodological value of combining data-driven estimation with a rule-based safety supervisor rather than replacing deterministic protection. The framework is intended as a transparent prototype for subsequent validation on public and laboratory lithium-ion datasets.