Machine Learning Health Estimation for Lithium-Ion Batteries Under Varied Conditions

To mitigate intermittency from renewable energy sources and present a sustainable alternative to fossil-fuel-based transportation, battery energy storage systems (BESSs) have drawn attention from both academia and industry in the last years. Despite different alternatives, Lithium-ion batteries (LIBs) have become the dominant technology for BESSs and electric vehicles. Therefore, knowledge of lithium-ion battery aging and lifetime estimation is a fundamental aspect for ensuring secure and reliable operations of different systems. This paper presents an analysis of five machine learning models, namely linear regression, k-nearest Neighbors (kNN), random forest (RF), support vector regression (SVR), multi-layer perceptron (MLP), in estimating the state of health (SOH) of LIB cells under different conditions. A total of 12 battery cells, cycled under three different temperatures ( \(15\,^{\circ }\textrm{C}\) , \(25\,^{\circ }\textrm{C}\) , \(35\,^{\circ }\textrm{C}\) ) and two discharge C-Rates (1C and 2C), were utilized for validation using mean absolute error (MAE) and R square ( \(R^2\) ) coefficient as performance indicators. Results indicated that both kNN and linear regression models achieved the lowest MAE values, with the linear regression model obtaining the highest \(R^2\) value. On the contrary, the MLP model showed the worse results among all models tested. A statistical analysis corroborated the results, indicating that the less complex learning models are suitable for estimating the non-linear SOH of LIBs.