Remaining Useful Life Prediction of Lithium-Ion Batteries With Limited Degradation History Using Random Forest

Predicting the remaining useful life (RUL) of a lithium-ion battery with its limited degradation history is critical as it ensures timely maintenance of electric vehicles (EVs) and efficient reuse of second-life batteries. Considering realistic battery operating conditions, this work investigates the RUL prediction under partial charge and discharge with a limited degradation history of the target cell. Given its ability to inform feature importance, the random forest is adopted to help prioritize different battery measurements and identify the least amount of operating data required for accurate RUL prediction. By examining the prediction performance using one complete charge and discharge cycle, it is shown that the duration, used capacity, and voltage signals of both charge and discharge contain important features related to battery RUL. The prediction performance under partial charge and discharge is also studied under state-of-charge (SOC) uncertainties, revealing satisfactory performance achieved with the data collected over the SOC range of $[{0.2,0.8}]$ . Comparison with an existing convolutional neural network (CNN)-based approach that uses four complete charge and discharge cycles verifies the enhanced onboard feasibility of the proposed approach. Sensitivity analysis against SOC ranges shows that the data in the SOC range of $[{0.1,0.2}]$ contain the richest RUL-related information for lithium iron phosphate (LFP) cells. Extensive validation on cells with different chemistry, ambient temperatures, and C rates further demonstrates the robustness of the proposed approach.