Degradation-Aware Federated Learning for Second-Life EV Batteries in Autonomous Nanogrids

Abstract

Repurposing retired electric vehicle (EV) batteries for stationary storage is a key pillar of the circular energy economy, but integrating second-life batteries (SLBs) into autonomous nanogrids is complicated by their heterogeneous degradation states and by the reluctance of automotive manufacturers to share proprietary cycling data needed to estimate state of health (SoH). We propose a decentralized diagnostic and control framework based on federated learning (FL), in which SoH-predictive neural networks are trained locally on edge controllers and only model weights are aggregated globally, followed by a degradation-aware dispatch algorithm that routes nanogrid power according to individual pack health. Numerical simulation shows the framework achieves 94.2% SoH prediction accuracy without sharing raw voltage or current data, while extending total nanogrid storage lifespan by 41% relative to standard equal-dispatch strategies.