Reduced Optimal Power Flow Using Graph Neural Network

TL;DR

This paper introduces a graph neural network-based method to reduce the complexity of optimal power flow problems, significantly decreasing computation time while maintaining solution accuracy in power system operations.

Contribution

The paper proposes a novel GNN approach to identify critical lines in power networks, enabling reduced OPF problems for faster real-time solutions.

Findings

GNN accurately predicts congested lines in power networks.

Reduced OPF achieves substantial computational savings.

Solution quality remains high despite problem reduction.

Abstract

OPF problems are formulated and solved for power system operations, especially for determining generation dispatch points in real-time. For large and complex power system networks with large numbers of variables and constraints, finding the optimal solution for real-time OPF in a timely manner requires a massive amount of computing power. This paper presents a new method to reduce the number of constraints in the original OPF problem using a graph neural network (GNN). GNN is an innovative machine learning model that utilizes features from nodes, edges, and network topology to maximize its performance. In this paper, we proposed a GNN model to predict which lines would be heavily loaded or congested with given load profiles and generation capacities. Only these critical lines will be monitored in an OPF problem, creating a reduced OPF (ROPF) problem. Significant saving in computing time is expected from the proposed ROPF model. A comprehensive analysis of predictions from the GNN model was also made. It is concluded that the application of GNN for ROPF is able to reduce computing time while retaining solution quality.