SIGS: A Synergistic Information-Guided Sparse Framework for Battery SOH Estimation

To address challenges in battery State of Health (SOH) estimation, which stem from the isolated optimization of feature engineering and model construction stages and the high computational cost of conventional methods, this paper introduces SIGS, a Synergistic Information-Guided Sparse framework. This multi-level collaborative framework first employs our novel Simplified Information Bottleneck (SIB) with a smoothness term to extract robust features and their importance scores. These scores then synergistically guide the subsequent stages: a new importance-weighted aging prototype selection (IW-APS) method identifies representative inducing points from the aging trajectory, and an importance-weighted kernel enhances the sparse Gaussian Process Regression’s focus on critical indicators. Comprehensive validation on four public datasets confirms the framework’s superior accuracy, with prediction root mean square error (RMSE) below 1% and outperforming state-of-the-art benchmarks across all scenarios. Furthermore, a Pareto analysis demonstrates its practical value by providing a clear guide for navigating the trade-off between prediction accuracy and computational cost.