Extended Long Short-Term Memory network for robust state-of-health estimation of lithium-ion batteries under diverse charging strategies

Accurate assessment of the health status of lithium-ion batteries is crucial for ensuring the safety and efficiency of their industrial applications. Various methods have been proposed to estimate the state of health (SOH) of lithium-ion batteries, but most of these methods are only applicable to specific types of batteries or operating conditions. To address this issue, this paper proposed an Extended Long Short-Term Memory (xLSTM) network for SOH estimation under various battery charging strategies. In this study, features are derived from the incremental capacity (IC) curve of the discharge process, which serve as primary indicators of battery health. To improve the accuracy of estimations, voltage and current features from both the charging and discharging phases are integrated as supplementary characteristics. The Spearman correlation coefficient is utilized to identify and select features that exhibit high correlation, thereby excluding irrelevant parameters. The proposed xLSTM model integrates the architectures of sLSTM and mLSTM, facilitating the effective capture of intricate temporal dependencies and nonlinear relationships associated with battery degradation. Experimental results demonstrate significant performance improvements, achieving average MAPE of 0.20% and R${}^2$ of 0.997, outperforming existing methods by over 40% in accuracy. Cross-chemistry validation on both LFP and NCM batteries confirms the robustness and generalization capability of the proposed method.