Adaptive personalized federated learning for the power Internet of Things
Hongbo Ma, Hai Chen, Wei Hua He, Xie Fang, Zifeng Zhang, Zhiyuan Yu · 2025
Federated Learning plays a crucial role in ensuring the security and privacy of data within the Power Internet of Things (PIoT) while enhancing the intelligence of energy management strategies. Addressing the issues of insufficient personalization and low communication efficiency in existing federated learning research on PIoT, we propose a personalized federated learning method with adaptive aggregation, implementing the Local Adaptive Personalized Aggregation (LAPA) algorithm. This approach comprehensively considers global knowledge alongside local personalization. Additionally, we introduce the dynamic adaptive parameter freezing algorithm, which finely adjusts the local aggregation weights and adaptively manages the range of changes in local client parameters through both unstructured and structured parameter freezing. Simulation results on authoritative public datasets demonstrate that our proposed method significantly improves model accuracy and enhances federated learning efficiency in heterogeneous scenarios typical of PIoT data.