Reconfigure Distribution Network with Physics-informed Graph Neural Network

Abstract

The reconfiguration of distribution networks is a complex problem that involves optimizing network topology to ensure efficient and reliable power delivery. Traditional approaches to this problem have relied on heuristics and optimization algorithms, which are computationally expensive and not scalable to large networks. In this paper, we propose a link prediction model based on a physics-informed graph neural network (GNN) by using the nodal and topological information of the distribution network. Numerical studies on a 119-bus distribution network show that the proposed physics-informed GNN exhibits a high level of accuracy in predicting the connectivity of tie lines. By synergistically combining the physics-informed GNN with an optimization model, the proposed algorithm significantly reduces the computation time of the network reconfiguration problem by using a subset of the link prediction results as the final tie switch connectivity.