Battery lifetime prediction considering domain-variate error

—With the rapid development of rechargeable battery technology, battery lifespan prediction has become a hot topic in current research. Data-driven models, due to their superior performance, have been widely applied in the field of battery lifespan prediction. These methods construct regression models by extracting features from early-cycle battery data to achieve accurate prediction of remaining useful life. However, non-ideal factors in real-world operating environments inevitably introduce noise interference into the raw battery datasets, and directly using noisy data for modeling significantly reduces prediction accuracy. To address this issue, the study proposes a noise-aware battery lifespan prediction framework based on modal decomposition. This framework employs a fully adaptive modal decomposition algorithm to decompose the original dataset, effectively removing noise components, and uses the high-quality derived features generated through interaction as inputs for the prediction model. Experimental validation on standard datasets demonstrates the framework's excellent predictive performance. To further evaluate the model's robustness, the study also conducted comparative experiments using a second dataset with added noise. The results show that, compared to traditional methods, the proposed approach exhibits significant noise reduction effects and notable improvements in prediction accuracy and stability, providing an effective solution for battery lifespan prediction.