Abstract
Large-scale retirement of lithium-ion batteries enables second-life use in decentralized energy communities, yet capacity-degradation prediction, residual-value assessment, and subsidy allocation are treated in isolation. This study proposes an integrated circular-economy framework linking battery-state prediction with community-level incentive coordination. A physics-constrained transfer-learning hybrid model estimates capacity trajectories and remaining useful life, achieving a root-mean-square error (RMSE) of 0.97%, R2 of 0.960, and cross-chemistry transfer RMSE of 1.05%. A dynamic residual-value mapping function and operational degradation-burden index characterize health losses. A Stackelberg leader-follower game is formulated and transformed into single-level optimization via KKT (Karush-Kuhn-Tucker) reformulation, with a dynamic subsidy-allocation mechanism redistributing equilibrium subsidies by degradation contribution, participation duration, and circular-economy factors. Experiments on 240 battery aging samples and a 60-prosumer community show the proposed mechanism yields welfare of 3,610 CNY/day, outperforming fixed uniform allocation under varying subsidy intensities and perturbations.