An ISO-LSSVM-AdaBoost Model for Estimating the State of Health of Lithium-Ion Batteries Over Their Entire Life Cycle

Abstract

Lithium-ion batteries undergo complex electrochemical reactions during charge-discharge cycles, significantly complicating the estimation of State of Health (SOH) throughout their entire lifecycle. To achieve precise SOH estimation under varying operating conditions, this paper proposes an ISO-LSSVM-AdaBoost model that integrates an improved snake optimization algorithm (ISO) with an adaptive boosting algorithm (AdaBoost). During data processing, the extracted health features (HF) are first denoised using a Variational Noise Adaptive H∞ Filter (VN-AHF). Subsequently, feature enhancement is applied to significantly improve their correlation with battery capacity decay. Enhanced features are quantitatively evaluated based on Pearson correlation coefficients, with the top 8 most correlated features selected as model inputs. Leveraging the ISO algorithm's strengths in parameter adaptive optimization and the robust regression capabilities of LSSVM-AdaBoost, this model achieves high-precision estimation of SOH throughout the battery's entire lifecycle. Experimental results demonstrate that the advanced ISO-LSSVM-AdaBoost model exhibits outstanding estimation performance under various operating conditions, including normal temperature, high temperature, and overcharging. Across the Oxford University dataset, Tsinghua University dataset, and Xi'an Jiaotong University dataset, the proposed ISO-LSSVM-AdaBoost method achieved a maximum root mean square error (RMSE) of 0.0036, with a mean absolute error not exceeding 0.0027, a minimum mean absolute percentage error of 0.0475% and the coefficient of determination consistently exceeded 0.99, validating its effectiveness and superiority as a battery SOH estimation method applicable over the life cycle under diverse conditions.

Keywords: Lithium-ion battery, state of health, Improved snake swarm optimization algorithm, Adaptive enhanced least squares support vector machine, Volume noise adaptive H-infinity filter