Enhanced Transformer-Long Short-Term Memory Network Modeling for Estimating Lithium-Ion Battery Health with Electrical Feature Extraction

As environmental pollution worsens, the use of clean energy becomes increasingly important, and the development of lithium-ion batteries is accelerating and growing in significance. The ability to accurately estimate the various states of lithium-ion batteries has become a trend in the development of battery management systems. This paper proposes a novel strategy for estimating the health status of lithium-ion batteries, based on publicly available datasets from NASA and the University of Maryland. Using the Kalman filter method, the IC features of lithium-ion batteries are extracted, and several feature factors with high correlation coefficients, such as peak values and peak areas, are selected. An improved hybrid model combining the Transformer model with the LSTM model is then used to accurately estimate the State of Health (SOH). The method is compared with LSTM, Transformer, BP, and CNN models, and the results show that it performs excellently in terms of prediction accuracy and robustness, achieving MAE, RMSE, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values of 0.0013, 0.0021, and 0.9817, respectively, significantly outperforming the independent models.