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

Area of Science:

• Battery Technology
• Artificial Intelligence
• Machine Learning

Background:

• Accurate prognostics of lithium-ion battery health are vital for electric transportation.
• Current data-driven methods require labeled data and domain-specific models, limiting real-world use.

Purpose of the Study:

• To explore the potential of large language models (LLMs) for generalized lithium-ion battery state-of-health forecasting.
• To evaluate LLM performance in few-shot and zero-shot learning scenarios for battery prognostics.

Main Methods:

• Utilizing LLMs in few-shot and zero-shot learning setups for battery health forecasting.
• Employing guided prompts to direct LLM predictions without model fine-tuning.
• Conducting extensive experiments to assess prediction accuracy under various conditions.

Main Results:

• LLMs demonstrated satisfactory performance in both few-shot and zero-shot learning.
• The lowest root-mean-square error achieved was [Formula: see text].
• Analysis revealed the impact of future operational conditions on prediction accuracy.

Conclusions:

• LLMs present a viable, generalized approach for lithium-ion battery health prognostics.
• Few-shot and zero-shot learning with LLMs can overcome limitations of traditional data-driven methods.
• This research highlights LLMs' potential for advancing battery management systems.