Battery State of Health Estimation for Real-world Vehicles Based on Ensemble Learning

Lithium-ion batteries (LIBs) are important energy storage components for new energy vehicles. Accurately estimating the State of Health (SOH) of LIBs ensures the safe and reliable operation of the power battery. Addressing the challenges of poor data quality, intricate sampling information, and incomplete battery cycles in real-world vehicle data, this paper proposes a pioneering SOH estimation method based on Light Gradient Boost Machine (LGBM) ensemble learning during charging process. A data preprocessing and operating condition reconstruction method is designed to streamline battery operating conditions and enhance overall data quality. The charging segments are extracted, and an aging features extraction method suitable for real-world vehicle data is proposed as well as high-value health indicators are selected using correlation analysis. The Bayesian optimization algorithm is employed to efficiently search for the optimal hyperparameters of the model, and comparisons and validations with other modeling algorithms and optimization methods and the estimation results on different vehicles are also carried out. The results show that LGBM method with Bayesian optimization algorithm achieves an overall superior estimation performance with the mean absolute percentage error limited within 2.71 % and exceptional computational efficiency.