A comprehensively optimized lithium-ion battery state-of-health estimator based on Local Coulomb Counting Curve

• Published in: Applied energy 2022-09, Vol.322, p.119469, Article 119469
• Main Authors: Huang, Huanyang, Meng, Jinhao, Wang, Yuhong, Feng, Fei, Cai, Lei, Peng, Jichang, Liu, Tianqi
• Format: Article
• Language: English
• Subjects: Comprehensive optimization; Data-driven model; Feature selection; State of health estimation
• ISSN: 0306-2619; 1872-9118
• DOI: 10.1016/j.apenergy.2022.119469

•Low complexity LCCC is proposed as features to be selected.•Unify the accuracy and feature acquisition difficulty into one objective function.•Feature selection and model training are simultaneously processed by GA.•Different variables in the gene are separately operated to avoid interference. Accurate State-of-Health (SOH) is critical to ensure the safe operation of Lithium-ion (Li-ion) batteries in electrified transportation and energy storage applications. The data-driven method is expected to greatly improve the SOH estimation in many aspects, thanks to the internet of things technology nowadays. Considering it is difficult to obtain valid information in real applications, efficient features and reasonable training procedures are two main points for establishing a superior data-driven SOH estimator. Thus, this paper proposes a comprehensive optimization framework for Li-ion battery SOH estimation with the Local Coulomb Counting Curve (LCCC), enabling both efficient feature extraction and good accuracy. Without the necessity of any complex calculations and smooth techniques, the LCCC in this work can be conveniently obtained by counting the coulomb amount of a specified voltage segment. After unifying the estimation accuracy and feature collection difficulty into one objective function, the Genetic Algorithm (GA) is utilized to optimize the LCCC selection and training procedure of the Gaussian Regression Process (GPR) further. Eight LiFePO4 batteries cycled under four different current rates aging conditions are selected for validation. The proposed estimator achieves root mean squared errors of 0.7745%, 1.0837%, 0.7208%, and 1.5795%, respectively, and optimized features can be collected within 300mV. Such results prove that the proposed method can achieve a good SOH estimation accuracy with fewer LCCC features and higher computing efficiency.