Causal Physics-Infused Hybrid Learning (CPIHL) Framework for Next-Gen Battery Health Forecasting

A novel hybrid model, denoted by Causal Physics-Informed Hybrid Learning Neural Networks (CPIHL), is developed in this study to significantly enhance the accuracy, interoperability, and real-time feasibility of battery health predictions. The model incorporates the effects of temperature and voltage on internal resistance, effectively capturing the non-linear electrochemical behaviors that drive battery degradation. The proposed framework is rigorously validated using two open-source datasets: the Samsung INR21700-50E and the Forklift Battery Degradation datasets. The CPIHL model demonstrates exceptional performance, achieving an R2 score of 0.9994, a mean absolute error of 0.0007, and a root mean square error of 0.0025, outperforming all baseline machine learning and deep learning models, including Random Forest, Artificial Neural Networks, Long Short-Term Memory, and Gated Recurrent Units. The CPIHL framework exhibits robust, interpretable, and scalable behavior, making it highly effective for predictive maintenance tasks. It provides actionable insights for battery management, enabling the optimization of operational strategies to extend battery life. By improving battery health monitoring, this work contributes to sustainable energy usage, enhancing efficiency and reducing battery disposal waste.