Physics‐Informed Neural Networks for Battery Degradation Prediction Under Random Walk Operations

This study addresses the challenge of predicting the state of health (SoH) and capacity degradation in Battery Energy Storage Systems (BESS) under highly variable conditions induced by frequent control adjustments. In environments where random walk behavior prevails due to stochastic control commands, conventional estimation methods often struggle to capture the degradation dynamics effectively. Our work presents a comprehensive approach that merges advanced feature engineering with state‐of‐the‐art machine learning techniques. Specifically, we introduce a physics‐informed neural network (PINN) model that incorporates domain‐specific physical laws—such as monotonic degradation and a first‐order decay law—into its training, thus limiting the hypothesis space to physically plausible functions. Comparative analysis with traditional methods such as Gaussian process (GP) regression, SVM, Lasso, and Ridge regression shows that the PINN model consistently delivers better predictive performance, evidenced by lower error metrics and improved values. These results underscore the potential of integrating physics‐based constraints with data‐driven models to enhance the reliability and accuracy of SoH and capacity degradation assessments in BESS, ultimately contributing to more effective battery management and control strategies.