Selective ensemble learning enables battery capacity estimation across charging rates

Accurate and robust capacity estimation is crucial for battery health management. Data-driven methods for battery capacity estimation have demonstrated significant application value. However, optimal models evaluated during the offline training phase exhibit substantial variations across different estimation cases. This leads to the unstable performance of single data-driven models on unseen test cases. Moreover, these models often require costly retraining when the battery's charging current changes. To address these issues, this paper proposes a selective ensemble learning (SEL) method for online capacity estimation across varying charging rates. Seven health features, extracted from segment charging data, indicate the battery's aging behavior under different charging currents. A non-negative least squares regression model is used as a meta-learner to selectively integrate eight heterogeneous machine learning models. Experimental results show that the proposed method achieves an average capacity estimation error of 0.99% across three batteries with different charging rates. Comparative results reveal that the proposed method outperforms the best single machine learning model by 12.4% and significantly exceeds several advanced ensemble learning methods and deep learning methods by over 10%. These results demonstrate that the proposed method can enhance battery capacity estimation performance while reducing ensemble complexity. This research highlights the substantial potential of applying selective ensemble learning to battery capacity estimation in complex charging scenarios.