Prompt-Driven Fine-Tuning of Large Language Model for Li-ion Battery State Estimation

Estimating the state of charge (SOC) and state of health (SOH) is essential for preserving the condition and improving the performance of Li-ion batteries. However, the inconsistencies between batteries and the diversity of working conditions greatly increase the difficulty of estimating SOC and SOH. Therefore, this paper proposes a prompt-driven fine-tuning of a large language model (LLM) for Li-ion Battery State Estimation. The method used to construct the LLM training dataset is a text encoder that combines the battery charge and discharge test data with the training text model. Using this dataset, we used prompt tuning to fine-tune the LLM to complete the battery state estimation task. The results show that the proposed method can achieve accurate state estimation in different working conditions and different types of batteries. Specifically, SOC estimates have a mean absolute error (MAE) of 0.44% under identical working conditions and a mean relative error (MRE) of 1.68%. The MAE and MRE of SOC estimates under different conditions are 2.21% and 4.73%, respectively. The SOH estimates yielded an MAE of 0.40% and an MRE of 0.48%.