Data-driven battery state-of-health estimation and prediction using IC based features and coupled model

Accurate estimation and prediction of the lithium-ion battery state of health (SOH) play a vital role in improving the reliability and safety of battery operations. However, the complexity of operation modes and inconsistency of aging trajectories in the real-world deteriorate the functional domain of existing methods in accurate estimation and prediction. In this study, a novel data-driven framework is proposed to enhance performance in real-world operation scenarios. Accordingly, an SOH estimation method is proposed, based on incremental capacity (IC) analysis and the operation characteristics of batteries. This method is more feasible in practical applications and has a 12.89 % improvement in reflecting the SOH compared with the IC peak. Moreover, a correction model is proposed and coupled with a regression model to remedy the deviation due to battery individual adaptively. The method is verified on laboratory and EV datasets, achieving mean absolute percentage errors of 0.29 % and 3.20 % respectively, evidently lower than those of conventional methods. This study highlights the adaptability of health features in real-world operation scenarios and the promise of combining group-based models with individual-based models to optimize predictions. The proposed framework can be extensively utilized for battery residual value analysis, secondary use, analysis of system energy storage, and other applications in real-world scenarios.

Automated summary

In this study, a novel data-driven framework is proposed to improve the accurate estimation and prediction of the state of health (SOH) of lithium-ion batteries. The proposed method, based on incremental capacity (IC) analysis and battery operation characteristics, is more suitable for practical applications and achieves a 12.89% improvement in reflecting SOH compared to the IC peak method. Additionally, a correction model is proposed to remedy deviations due to battery individual adaptivity. The method is validated on laboratory and EV datasets, showing significantly lower mean absolute percentage errors than conventional methods. The study highlights the adaptability of health features in real-world scenarios and the potential of combining group-based and individual-based models for optimized predictions.