Capacity Degradation Estimation of Lithium-Ion Batteries Using Explainable Machine Learning
TL;DR
This research presents a machine learning approach to predict the capacity degradation of lithium-Ion battery, integrated with Explainable AI to improve model explainability and shows that charge time has the most impact on capacity degradation prediction.
Abstract
Electric vehicles have emerged as a sustainable solution to reduce carbon footprint in the transport sector. However, a major challenge faced by EVs is their battery degradation. Among various battery's health indicators, capacity degradation serves as a crucial metric to quantify battery health. This research presents a machine learning approach to predict the capacity degradation of lithium-Ion battery, integrated with Explainable AI to improve model explainability. Multiple supervised machine learning models including CatBoost, Extra Trees, LightGBM and Bagging Regressor were implemented and evaluated using performance metrics to determine the best-performing model. Furthermore, the Explainable AI (XAI) technique, Shapely Additive Explainable (SHAP), was applied to the best-performing model for explainability and to show the contribution of different features on capacity degradation prediction. The methods used resulted in $R^2$ value of 0.974 using the CatBoost model and the information provided by SHAP show that charge time has the most impact on capacity degradation prediction.
