Lithium-Ion Battery State of Health Estimation Based on LSTM-Transformer

Deep learning methods have demonstrated potential in estimating the health state of lithium batteries, which is essential for safety management. Yet, they often struggle to capture long-term dependencies and global correlations within battery capacity sequences. To address this challenge, a fused LSTM-Transformer approach for lithium battery health state estimation is proposed. the excellent sequence modelling capability of LSTM enables it to effectively capture the long-term dependence of battery performance evolution over time and adapt to pattern changes in different time scales, while the Transformer model achieves excellent global correlation capture through the self-attention mechanism. to understand the charging and discharging behaviour of batteries in a more global perspective. This integration allows the models to surpass the limitations inherent in using a single model and improves the ability to model the dynamic complexity of battery systems. Additionally, experimental results using a public dataset confirm that the method introduced in this paper offers a more thorough and precise evaluation of battery health compared to alternative deep learning artifacts.