GPT4Battery: Cross-battery State of Health Estimation via Physical-Guided Test-time Prompt Learning with LLM
Published in Arxiv, Under Review, 2024
State of health (SOH) is a crucial indicator to evaluate the level of degradation of batteries that cannot be measured directly but requires estimation. Accurate SOH estimation enhances detection, control, and feedback for Li-ion batteries, allowing for safe and efficient energy management and guiding the development of new-generation batteries. Despite significant progress in data-driven SOH estimation methods, the time- and resource-consuming degradation experiments to generate lifelong training data pose a barrier to timely safety monitoring and the development of new battery technologies. To tackle this problem, We propose GPT4Battery, the first foundation model for battery to utilize the strong generalization ability of the large language model (LLM) for cross-battery SOH estimation. We design a translator to reprogram the voltage-time battery charging data into text prototype representations to align data from different modalities, adapting the LLM to battery tasks. We also design a novel physical guided test-time prompt tuning (PGTPT) method to learn adaptive prompts on the fly with a single test sample, enhancing the model generalization ability and fitting the real-world scenarios. PGTPT optimizes the prompt by guiding the LLM to generate a complete battery charging curve in accordance with the physics equations of the 1-RC ECM model of a LIB cell. The validation results demonstrate that the proposed model achieves state-of-the-art accuracy that is even on par with domain adaptation or fine-tuning methods that require additional training data on five widely recognized datasets collected from 65 batteries.
Recommended citation: To be announced. https://github.com/yuyuan6/yuyuan6.github.io/raw/master/files/gpt4battery.pdf