State of Health Estimation for Lithium-Ion Batteries via Fused Impedance Feature and V-Transformer-AGLU Network

Accurate State of Health (SoH) estimation is critical to ensuring the safety and optimizing the performance of lithium-ion batteries. To address the limitations of existing methods in capturing the complex aging mechanisms and long-range temporal dependencies, this paper proposes a novel SoH estimation approach for lithium-ion batteries that integrates the Fused Impedance Feature (FIF) with a Vision Transformer embedded with an Adaptive Gated Linear Unit (V-Transformer-AGLU). Specifically, the FIF is constructed by fusing the Mean and Median of dynamic impedance, calculated from individual cells within the SoC range of 0.3–0.7. This fusion strategy explicitly balances the sensitivity of the Mean to global degradation trends with the robustness of the Median to local anomalies and noise. Furthermore, FIF exhibits particular effectiveness in modeling the SoH for series-connected battery packs with cell inconsistency. Subsequently, an efficient V-Transformer-AGLU model is constructed, which marks the first application of the V-Transformer to one-dimensional temporal SoH prediction for battery systems. The model adopts a patching mechanism to capture global dependencies, and AGLU is employed to substitute the conventional Feedforward network for facilitating dynamic information fusion and enhancing robustness against the nonstationary characteristics of battery signals. Experimental results verify the high accuracy and generalization robustness of the proposed method, which achieves a root-mean-square error (RMSE) of 0.0054, a mean absolute error (MAE) of 0.0043, and a coefficient of determination (R2) of 0.9868 on the 8-cell pack and maintains outstanding performance across different packs. These results provide valuable technical support for the online monitoring of battery management systems (BMS).