Leveraging Power Grid Topology in Machine Learning Assisted Optimal Power Flow

Machine learning assisted optimal power flow (OPF) aims to reduce the computational complexity of these non-linear and non-convex constrained optimization problems by consigning expensive (online) offline training. The majority work in this area typically employs fully connected neural networks (FCNN). However, recently convolutional (CNN) graph (GNN) have also been investigated, effort exploit topological information within grid. Although promising results obtained, there lacks a systematic comparison between architectures throughout literature. Accordingly, we introduce concise framework for generalizing methods machine OPF assess performance variety FCNN, CNN GNN models two fundamental approaches domain: regression (predicting generator set-points) classification active set constraints). For several synthetic grids with interconnected utilities, show that locality properties feature target variables are scarce subsequently demonstrate marginal utility applying compared FCNN fixed grid topology. variable topology (for instance, modeling transmission line contingency), able straightforwardly take change into account outperform both models.