State of Health Estimation for Lithium-Ion Batteries Using Physics-Informed Health Factors and Cascaded Optimized HKELM

Abstract Accurate state of health (SOH) estimation is essential for ensuring the safety, reliability, and economic operation of lithium-ion battery systems. However, the battery aging process is highly nonlinear and strongly affected by feature quality, parameter optimization efficiency, and model generalization under limited samples. To address these issues, this paper proposes a data-driven SOH estimation framework based on physics-informed health factor extraction and a cascaded optimized hybrid kernel extreme learning machine, termed C-HBA-IGWO-HKELM. First, four health factors with clear physical relevance to battery degradation, including constant-current charging time, constant-voltage charging time, the peak value of the incremental capacity curve, and the corresponding peak voltage, are extracted from the charging process to characterize battery aging behavior. Second, a cascaded optimization strategy is developed by combining circle chaotic initialization, a honey badger algorithm-based global exploration mechanism, and an improved grey wolf optimizer-based local exploitation mechanism, so as to enhance parameter search capability and convergence stability. Third, a hybrid kernel extreme learning machine with an adjustable regularization term is constructed by combining polynomial and radial basis kernels to improve nonlinear fitting ability and reduce the risk of overfitting. Experiments conducted on the NASA lithium-ion battery dataset show that the proposed model achieves better overall performance than several comparison models in terms of R², RMSE, MAE, and MAPE. In addition, sensitivity analysis under different training-test split ratios further demonstrates the stability and adaptability of the proposed framework. The results indicate that the proposed method provides an effective and computationally efficient solution for lithium-ion battery SOH estimation.