.png){kind=icon}
.png){kind=icon}
Return
Li-Ion Battery Doctor: A Fine-Tuned and Explainable Large-Language Model for Health Prognosis
.png){kind=icon}
PRE
AI
R
L
徐
DOI：10.1109/TEC.2026.3651941.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Health prognosis is crucial for reliable operation of Li-Ion batteries (LIB). Recently, machine learning (ML) methods have attracted significant interest for state-of-health (SOH) prediction of LIB. However, most ML methods overlook the impacts of varying operational conditions on battery SOH, and their generalization ability and interpretability are usually concerned. With the recent advances of large language models (LLMs), this letter proposes an LLM based framework for health prognosis of LIB, which contains two stages: offline parameter-efficient fine-tuning and online applications with retrieval-augmented-generation (RAG) system integration. The first stage fine-tunes the LLM using low-rank adaptation for efficient training with lower memory and computational costs, making it ideal for adapting large models to specific tasks with limited computational resources. In the second stage, the trained model is applied with integrated domain knowledge to enhance explainability. The RAG system interprets the underlying principles of degradation using its knowledge base, providing guidance to adjust operational conditions for improving battery health. The proposed framework is implemented with distilled DeepSeek-r1 models and achieves high prediction accuracy. Additionally, the RAG system provides credible and context-aware explanations for prognosis results.
Keywords:
Large language model
parameter-efficient fine-tuning
RAG
Li-ion battery
health prognosis
Journal
IF:
5.4
Papers:
6.8K
Citations:
1.5W
Organization
Cited Papers