Power system transient stability assessment via physics-informed graph convolutional networks

Abstract

Transient Stability Assessment is the first line of defense for the safe and stable operation of power grids, which is decisive for ensuring the continuity of power supply and reducing system operational risks. Addressing the challenges that traditional time-domain simulation methods have low computational efficiency and are difficult to meet real-time requirements, and that purely data-driven Graph Convolutional Networks (GCN) lack physical interpretability and their prediction results easily violate physical laws, this paper proposes a new TSA method fusing physical mechanism constraints and GCN. Firstly, combining the grid topology and electrical parameters, a graph model fusing physical characteristics is constructed, and key electrical physical quantities are selected as node features. Secondly, a Physics- Informed Graph Convolutional Network is designed. By deeply embedding physical laws such as network topology constraints and generator rotor motion equations into the graph convolution aggregation mechanism and loss function optimization process, the effective fusion of data-driven and physical mechanisms is achieved. Finally, simulation results on IEEE 39-bus and 118-bus standard test systems show that the proposed method achieves millisecond-level rapid assessment, with accuracy and instability recall rates both exceeding 97%. While ensuring assessment accuracy, it significantly improves the physical consistency of the model, providing strong technical support for online security early warning and emergency control decisions of modern power grids.