State of health estimation for lithium-ion batteries via electro-thermal feature fusion and a GRU-Guided Attention Network

Critical for battery safety though accurate State of Health prediction is, existing models lack joint capture of local dynamics and global degradation trends. We propose a novel GRU-Guided Cross-Attention Network (GCA-Net) leveraging multi-physics feature fusion. The architecture employs GRU layers to model long-range degradation dependencies, guiding an asymmetric cross-attention module to adaptively focus on key multi-physics features extracted from Incremental Capacity Analysis (ICA) and Differential Thermal Voltammetry (DTV) curves. Evaluations on the NASA and MIT public benchmark datasets demonstrate high prediction accuracy and strong robustness against controlled noise and thermal perturbations. • GRU-Guided Cross-Attention Network (GCA-Net) is proposed for battery SOH estimation. • Adaptive IC/DTV health feature extraction tailored to different datasets. • GRU-Guided asymmetric attention adaptively focuses on key degradation features. • Robust against random noise and temperature disturbances.