Enhanced state of charge estimation for solid-state batteries using a stacked ensemble machine learning model

As the demand for higher-performing batteries intensifies, particularly in terms of safety, energy density, and cycle life, solid-state batteries have gained prominence due to their stable solid electrolytes, which eliminate leakage and explosion risks associated with liquid electrolytes. To ensure the safe and efficient operation of batteries, accurate State of Charge (SOC) estimation is critical within Battery Management Systems (BMS). This paper proposes a novel SOC estimation approach based on supervised machine learning, using a stacked ensemble model that integrates Random Forest (RF), Gaussian Process Regression (GPR), and Gradient Boosted Decision Trees (GBDT) as base learners. The model combines predictions from these learners with the original input features to train a meta-learner, enhancing the system's ability to capture the nonlinear behavior of battery SOC. Feature selection, Bayesian hyperparameter optimization, and cross-validation are applied to improve robustness and generalizability. Experimental results on a dataset of over 17,000 charge–discharge cycles demonstrate that the proposed model achieves an RMSE of 0.73% and MAE of 0.61% at 25 °C, representing a 43% improvement in RMSE compared to the best-performing single model. The model maintains high accuracy even in challenging SOC regions (< 20% and > 80%), making it a reliable and scalable solution for solid-state battery management.