Clustered Federated Learning for Energy-Harvesting Smart Meters in P2P Energy Trading
Ziming Liu, Bonan Huang, Cheng Zhang, Zhongming Yao, Tianyi Li, Qiuye Sun, Yushuai Li
Abstract
Energy-harvesting smart meters enable decentralized integration of renewable energy through peer-to-peer (P2P) trading. However, they face critical challenges, e.g., limited transaction participation, fluctuating communication reliability, and accelerated battery degradation. To address these issues, we propose CAFL, a blockchain-based Clustered Asynchronous Federated Learning framework tailored for smart meter communication networks. First, we propose an energy-aware FL protocol that dynamically prioritizes devices with sufficient energy reserves and stable channel conditions, effectively decoupling model training continuity from the non-stationary nature of energy-harvesting communications. Second, we design an adaptive dual-phase clustering strategy that integrates gradient similarity analysis with adaptive differential privacy, mitigating non-independent and identically distributed data skewness while safeguarding against preference inference attacks under intermittent connectivity. Third, we propose a blockchain architecture that establishes tamper-proof associations between FL contributions and P2P trading rewards through a contribution-weighted consensus mechanism. With those efforts, CAFL enables synergistic optimization of energy-harvesting allocation, distributed learning efficiency, and privacy-energy tradeoffs. Simulation results validate the effectiveness of our approach.