Scholarlink

<p>Prognostics of the health degradation of lithium-ion batteries plays a crucial role in the electrification of transportation systems. Data-driven approaches have been widely adopted for battery health forecasting, but the need for labeled data and the domain-specific nature of the prediction model limit the deployment of these approaches in real-world applications. In this study, we unlock the potential of large language models as a generalized approach for lithium-ion battery state-of-health forecasting. We demonstrate the feasibility of applying large language models in a few-shot and zero-shot learning setups, where the model is capable of forecasting the battery health degradation given proper guided prompts without the need for fine-tuning. Extensive experiments are conducted to evaluate the prediction performance with various usage setups considered. The results indicate that both few-shot and zero-shot learning setups yield satisfactory performance with the lowest root-mean-square error of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(1.34\%\)</EquationSource> </InlineEquation> achieved. In addition, this research examines various future operational conditions provided in the prompt and their impact on the prediction performance. The findings of this study provide insights into the potential of large language models as a generalized approach for lithium-ion battery health prognostics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prognostics by generalists: large language models for lithium-ion batteries health forecasting

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月9日

查看全文
  • Kei Long Wong,
  • Sio Kei Im,
  • Xinyi Fang,
  • Su-Kit Tang,
  • Chan Tong Lam,
  • Giovanni Pau

摘要

Prognostics of the health degradation of lithium-ion batteries plays a crucial role in the electrification of transportation systems. Data-driven approaches have been widely adopted for battery health forecasting, but the need for labeled data and the domain-specific nature of the prediction model limit the deployment of these approaches in real-world applications. In this study, we unlock the potential of large language models as a generalized approach for lithium-ion battery state-of-health forecasting. We demonstrate the feasibility of applying large language models in a few-shot and zero-shot learning setups, where the model is capable of forecasting the battery health degradation given proper guided prompts without the need for fine-tuning. Extensive experiments are conducted to evaluate the prediction performance with various usage setups considered. The results indicate that both few-shot and zero-shot learning setups yield satisfactory performance with the lowest root-mean-square error of \(1.34\%\) achieved. In addition, this research examines various future operational conditions provided in the prompt and their impact on the prediction performance. The findings of this study provide insights into the potential of large language models as a generalized approach for lithium-ion battery health prognostics.