Feature Separation of Health Factors for Domain-Adaptive Battery Capacity Estimation

Accurate capacity estimation is essential for the operational safety of lithium-ion batteries. Traditional domain-adaptive methods, while aligning features across different domains, often overlook important domain-specific features, which can lead to reduced predictive accuracy. To address this issue, we propose the Health Factor Feature Separation Network (HFFS-Net), a novel approach designed to separate hidden representations of handcrafted health factors into domain-invariant and domain-specific features. This strategic separation preserves critical degradation information, improving adaptability across various usage conditions. Comparative experiments show that HFFS-Net outperforms existing state-of-the-art methods, delivering superior accuracy in capacity estimation, which is crucial for ensuring battery safety.