Lag-Llama-based Remaining Useful Life Prediction for Lithium-ion Batteries of Electric Vehicles with Auto-Correlation Analysis

The accurate prediction of the remaining useful life of lithium-ion batteries is essential for applications like electric vehicles, but current methods often struggle to generalize across different battery types and operational conditions due to their reliance on large labeled datasets. This paper proposes a novel approach that integrates the Lag-Llama foundation model with auto-correlation analysis for remaining useful life prediction. Firstly, the tokenization scheme of Lag-Llama is enhanced using auto-correlation to identify degradation patterns in historical capacity sequences, improving the model’s ability to capture capacity fluctuations. The model is pre-trained on the large RWTH Aachen University dataset to learn degradation trends and fine-tuned using smaller, target-specific datasets, including the Center for Advanced Life Cycle Engineering (CALCE), National Aeronautics and Space Administration (NASA), Massachusetts Institute of Technology (MIT), and Battery Analytics with Artificial Intelligence (BAWAII) datasets. Compared to the baselines, the proposed method reduces mean absolute errors by 43.2% on CALCE, 55.2% on NASA, 78.9% on MIT, and 91.1% on BAWAII. In terms of relative error, improvements of 48.7%, 54.9%, 80.2%, and 91.2% are achieved on the same datasets, respectively. These results demonstrate the model’s superior accuracy and generalizability for data-driven remaining useful life prediction of lithium-ion batteries, even with limited fine-tuning data.