An Optimized Random Forest Regression Model for Li-Ion Battery Prognostics and Health Management

Abstract

This study proposes an optimized random forest regression model to achieve online battery prognostics and health management. To estimate the battery state of health (SOH), two aging features (AFs) are extracted based on the incremental capacity curve (ICC) to quantify capacity degradation, further analyzed through Pearson’s correlation coefficient. To further predict the remaining useful life (RUL), the online AFs are extrapolated to predict the degradation trends through the closed-loop least square method. To capture the underlying relationship between AFs and capacity, a random forest regression model is developed; meanwhile, the hyperparameters are determined using Bayesian optimization (BO) to enhance the learning and generalization ability. The method of co-simulation using MATLAB and LabVIEW is introduced to develop a battery management system (BMS) for online verification of the proposed method. Based on the open-access battery aging datasets, the results for the mean error of estimated SOH is 1.8152% and the predicted RUL is 32 cycles, which is better than some common methods.