Power system preventive control aided by a graph neural network-based transient security assessment surrogate

Abstract

In the context of the clean energy revolution and the high penetration of renewables and power electronics, the uncertainty level of operating states significantly increases, which brings new challenges to the safe and stable operation of power grids. Power system preventive control is an important measure to ensure the safe and stable operation of power systems by adjusting the active generation and nodal voltage of generators. In this paper, a power system preventive control algorithm aided by a Graph Neural Network (GNN)-based Transient Security Assessment (TSA) surrogate is proposed. A GNN-based fast contingency scanning model is constructed based on power network topology, which can predict the contingency scanning results rapidly only based on the power flow by transforming the original contingency scanning problem into the node classification problem on the GNN. The obtained GNN is then used as a surrogate model for TSA to speed up the solution of particle swarm optimization-based transient security-constrained optimal power flow. Numerical Tests are carried out in the IEEE-39 system and the test results indicate that the proposed method can significantly improve the efficiency without affecting the solution accuracy.