State-of-Charge Prediction of Degrading Li-ion Batteries Using an Adaptive Machine Learning Approach

Abstract

State of charge (SOC) estimation of degrading batteries is important for battery energy storage systems (BESS) employed in power system applications and electric vehicles. This paper aims to propose a comparative analysis for data-driven models such as linear regression (LR), support vector regression (SVR), random forest (RF), and Gaussian process regression (GPR) to estimate the battery SOC at various temperatures and loading levels considering the state of health (SOH) of the battery. The historical data in which the cells are degraded from SOH of 100% to 60% are employed to extract the correlated features with the SOC. The models are retrained and adaptively updated based on the new SOH and prepared to estimate the SOC at the current SOH. The results demonstrate that GPR and RF models have the best performance. The mean absolute error of less than 0.0223 and 0.0204 have been achieved for RF and GPR, respectively.