Optimizing charge discharge cycles using QPPONet-enabled hybrid learning framework for energy management and safety in electric vehicles.

The move toward the use of electric vehicles has underscored some serious 3.4s in traditional Battery Management System (BMS), which highly relies on the rule-based logic and empirical formulations. These systems fail to cope with these changing battery conditions, which usually leads to inaccurate health forecasts, poor energy consumption, and high safety hazards. This study suggests an adaptive, intelligence-driven BMS architecture, which makes use of machine learning and reinforcement learning algorithms to address the following intrinsic constraints. In the case of State of Health (SOH) estimation, the Gradient Boosting model performed best, having almost perfect accuracy in prediction on both proprietary and open-source datasets with a range of R2 at 0.9926–0.9995, which is much better than other models, namely, Random Forest, Extreme Gradient Boosting, and Long Short-Term Memory (LSTM) networks. A hybrid LSTM-GRU model in Remaining Useful Life (RUL) prediction had a Root Mean Square Error (RMSE) of 3.36 and a Mean Absolute Error (MAE) of 2.09 and was able to better predict long-term dynamics of degradation than a standalone recurrent model. The hybrid QPPONet (DQN and PPO) reinforcement learning approach led to a 24 ± 1.2% gain in cumulative reward while making balanced efficiency and degradation management trade-offs. Meanwhile, system safety was enhanced by means of unsupervised anomaly detection with the Isolation Forest and One-Class SVM algorithms, which delivered more than 92% accuracy in terms of early thermal runaway detection. Instead of putting all modules together into one pipeline, each component was verified separately to allow modular, scalable integration into a wide range of hardware and deployment environments. The framework was tested on a proprietary high-resolution EV dataset and on the publicly available eVTOL dataset, showing strong generalization across heterogeneous cycling conditions. These results establish a scalable roadmap toward intelligent, secure, and adaptive BMS architectures, thereby informing the broader pursuit of reliable, sustainable electric mobility.