Towards the Adoption of Large Language Models for Interactive Digital Twin in Battery Management Systems

초록

In this paper, we propose a novel digital twin assistant for battery management systems (BMS) to interpret the insights the digital twin provides. Our proposed digital twin assistant is based on a FLAN-T5 large language model (LLM), fine-tuned using a low-rank adaptation parameter-efficient fine-tuning approach for more domain-specific responses. The model was trained on a dataset specifically derived from a NASA battery dataset and was tailored to provide prompts and responses for training and testing the FLAN-T5 model. Experiment results show, the fine-tuned FLAN-T5 model achieves a rogue1, rogue2, rogueL, and rogueLsum scores of 0.400, 0.152, 0.286, and 0.292 respectively. These results highlight the potential of the proposed approach in ensuring a more explainable and user-friendly digital twin for BMS.