A novel state of health estimation method for lithium-ion battery based on forward-broad learning system

Abstract

Accurate battery state estimation is crucial for optimizing performance, enhancing safety, and prolonging battery life. To improve predictive accuracy of State of Health (SOH) and enhance the generalization capability, this paper proposes a novel framework for SOH estimation based on Forward-Broad Learning System (F-BLS). Firstly, health features are extracted from the charging data, and correlation analysis is conducted to select health metrics highly correlated with battery life degradation. Secondly, in order to achieve efficient training of the model, an adaptive parameter optimization algorithm Forward-AdaBound (FAdaBound) is, integrated into the Broad Learning System (BLS) to create the F-BLS. In addition, the proposed F-BLS integrates regularization techniques to improve its generalization performance. Finally, three datasets are employed to evaluate the performance of the proposed method, which is compared it with two commonly used data-driven methods. The test result demonstrates that the proposed SOH estimation method accurately tracks the capacity degradation of the battery, with RMSE less than 0.02 and MAE less than 0.03. Importantly, the F-BLS not only achieves excellent training metrics but also sustains good prediction accuracy on unforeseen test samples, showcasing strong generalization capability.