AI-Driven Dynamic Energy Routing in P2P Energy Systems

Abstract

The rapid evolution of smart grids and the increasing integration of renewable en-ergy resources have transformed traditional electrical networks into decentralized and intelligent energy systems. Peer-to-peer (P2P) energy trading has emerged as a promis-ing paradigm that enables direct energy exchange between distributed producers and consumers, improving renewable energy utilization and reducing dependency on cen-tralized infrastructures. However, efficient energy routing in P2P systems remains a challenging problem due to network complexity, dynamic energy generation, variable demand, and transmission losses. This thesis proposes an artificial intelligence-based approach for dynamic energy routing and power loss prediction in P2P energy systems using Graph Neural Networks (GNNs). The electrical distribution network is mod-eled as a graph structure where nodes represent households or prosumers and edges represent physical electrical connections. Node and edge attributes, including energy production, consumption, distance, resistance, and network characteristics, are inte-grated into the learning process. A complete data processing pipeline is developed to transform raw energy consumption and solar generation data into realistic P2P energy transactions with associated physical constraints. A physics-aware Graph Attention Network (GAT) model is designed to learn complex spatial relationships between net-work components and predict energy routing losses while selecting efficient transmission paths. Experimental results demonstrate that the proposed model achieves accurate power loss prediction and improves routing efficiency compared with traditional ap-proaches. The obtained results highlight the potential of graph-based deep learning techniques as scalable solutions for future intelligent energy networks