Fuzzy-Logic-Based Novel Aggregation Method for Federated Learning: Application to a Solar PV Energy Generation System

Federated learning enables collaborative model training across decentralized devices, preserving data privacy by keeping sensitive information localized. Federated learning facilitates the development of robust machine learning models by leveraging diverse datasets distributed across numerous clients, without necessitating data centralization. Federated learning frameworks utilize a suite of aggregation algorithms, such as weighted averaging (FedAvg), proximal optimization (FedProx), and adaptive optimization (FedOpt), to perform a convergent synthesis of locally computed model parameter vectors, thus enabling distributed model refinement while maintaining data locality. Fuzzy logic provides a computational framework for approximate reasoning, enabling the modeling of imprecise and uncertain information through the utilization of fuzzy sets and linguistic variables. This study introduces the development of a novel fuzzy-logic-driven aggregation mechanism, FedFZY, for a federated learning approach. This implemented methodology eliminates the requirement for complex mathematical operations and derivative applications. To validate the efficacy and observe the operational behavior of the proposed methodology, the FedFZY method has been deployed on a photovoltaic solar energy generation system, comprising 14 clients. The obtained results have been subjected to comparative analysis with alternative aggregation methodologies, yielding demonstrably successful outcomes.