Accurate remaining useful life estimation of lithium-ion batteries in electric vehicles based on a measurable feature-based approach with explainable AI

As Electric Vehicles (EVs) become increasingly prevalent, accurately estimating Lithium-ion Batteries (LIBs) Remaining Useful Life (RUL) is crucial for ensuring safety and avoiding operational risks beyond their service life threshold. However, directly measuring battery capacity during EV operation is challenging. In this paper, we propose a novel approach that leverages measurable features based on the discharge time and battery temperature to estimate RUL. Our framework relies on a novel feature extraction strategy that accurately characterizes the battery, leading to improved RUL predictions. Multiple machine learning algorithms are employed and evaluated. Our experimental results demonstrate that the proposed method accurately estimates capacity with minimal hyperparameter tuning. The R2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R^2$$\end{document} scores across various battery numbers indicate strong predictive performance for models like XGBoost, RF, AdaBoost, and others, with improvement percentages ranging from 85% to 99%, which the model’s generalizability verifies across other batteries. The results show the effectiveness of our proposed method in accurately estimating the RUL of LIBs in EVs.