HML-BattNet: A Hybrid Machine Learning Framework for Real-Time Battery Health Prediction in Electric Vehicles

The rapid advancement of electric vehicles (EVs) has increased the demand for reliable and real-time battery health management systems to ensure safety, efficiency, and prolonged battery lifespan. Most battery health prediction methods which rely on physics or simple machine learning, have little ability to change with new situations, recognize various driving patterns well or respond quickly enough in changing situations. To address these issues, we introduce HML-BattNet which includes 1D Convolutional Neural Networks (1D-CNNs), Bidirectional Long Short-Term Memory (BiLSTM) networks and Gradient Boosting Decision Trees (GBDTs) for accurate battery SoH estimation. The model is created using extensive data from various voltage, current, temperature and state-of-charge (SoC) measurements in various operating situations. Over 10-fold cross-validation, it is shown that HML-BattNet is better than CNN-LSTM, BiLSTM and Random Forest in all important measures. The model has an average accuracy of 96.1%, an F1-score of 0.956 and a Mean Absolute Error (MAE) of 0.0275, all of which make it suitable for quick processing on vehicles. Also, the framework facilitates the use of domain adaptation, making it possible to adapt with different battery chemistries using transfer learning. What this demonstrates is that HML-BattNet is flexible, can be used widely and adapts quickly to different EV battery management systems. BattNet stands out from other models due to using deep temporal feature extraction and an easy-to-understand gradient boosting logic. This work makes it possible to develop better ways to look after battery safety and health.