A Federated Learning Framework for Proactive Congestion Management in Blockchain-Based P2P Energy Markets

Abstract

The rapid penetration of distributed energy resources (DERs) is transforming consumers into prosumers, enabling decentralized peer-to-peer (P2P) energy markets. However, a major obstacle to their practical use is making sure that trades based on economics don’t break the physical limits of the distribution grid. This paper presents an innovative, entirely decentralized, and contextually aware Federated P2P (FP2P) market framework to tackle this issue. The framework co-optimizes the well-being of all prosumers with the safety of the grid by putting a linearized ac power flow model directly into a blockchain-based market clearing system. The main new thing about it is a proactive congestion management system that uses a Federated Learning (FL) architecture that protects privacy. It uses a Graph Neural Network (GNN) to predict the likelihood of line congestion. This predicted risk is turned into a dynamic grid fee, which keeps the market from going to unsafe operating points. The proposed FP2P framework was rigorously validated through a high-fidelity co-simulation on a modified IEEE 37-bus test feeder using real-world prosumer, weather, and market data. Results demonstrate that our approach successfully eliminates grid congestion events, significantly increases total prosumer profit, and reduces peak substation load compared to centralized, unmanaged P2P, and rule-based control baselines. The complete, AI-driven solution this work provides that bridges the gap between market economics and grid physics will enable future decentralized energy markets to function in a safe, efficient, and scalable way. © 2013 IEEE.