Data-driven remaining useful life (RUL) prediction depends on large, high-quality datasets. To address data scarcity, we propose a meta-learning framework for fast-adaptive early-stage RUL prediction of lithium-ion batteries, exploiting only a limited number of degradation profiles to learn a transferable initialization that can quickly adapt to unseen cells and supports robust recursive multi-step forecasting.
Test results show that the proposed framework thrives under scarce degradation data and rapidly adapts to unseen cells, outperforming baseline and transfer learning models in most scenarios, particularly for long-horizon recursive prediction covering the deep-degradation regime with an MAE of 1.156%. In-depth analyses are also presented regarding the influence of the prediction horizon and history length.