State of Health Estimation for Lithium-Ion Batteries Based on Global Feature Extraction and KPCA-ISSA-LSSVM Model

Accurate assessment of lithium-ion batteries (LIBs) state of health (SOH) is crucial for ensuring the safety of electrochemical energy storage systems and electric vehicles, as well as enhancing the reliability of battery management systems. However, reliance on single health feature extraction and suboptimal model selection increases computational burden and compromises estimation accuracy. To overcome these limitations, this study proposes a novel SOH estimation framework for LIBs that integrates comprehensive feature extraction with an optimized hybrid model—KPCA–ISSA–LSSVM to improve both prediction accuracy and efficiency. Sixteen HFs are extracted from charge–discharge segments of different cycles and categorized into four groups—current, voltage, temperature, and incremental capacity—to characterize electrochemical aging mechanisms underlying capacity degradation. After Gaussian filtering and anomaly removal, dual-correlation analysis with normalized curve assessment selects features with correlation coefficients above 0.8, enhancing feature quality. KPCA then reduces the dimensionality of these correlated features before input into the ISSA–LSSVM predictor, improving model training efficiency and accuracy while maintaining low complexity. Comparative experiments on three NASA battery datasets show that the proposed method achieves mean absolute errors below 0.6% and R 2 above 0.997 across different training ratios, significantly outperforming benchmark models in prediction accuracy and robustness.