基于图神经网络的电力系统实时多稳定性风险评估与可视化

Multi-stability risk assessment (MSRA) is more practical than singular stability risk assessment in power system operation considering increasing uncertainties, e.g., renewable power generation and system faults. In this paper, a real-time MSRA method based on a graph neural network (GNN) is proposed to effectively address multiple stability problems, including (small-disturbance and transient) rotor angle, (short-term and long-term) voltage, frequency, and converter-driven stability. An operating graph and a disturbance graph are developed as input features of GNN to completely characterize complex operating conditions and disturbances. In the GNN, the topology correlations in the inputs can be learned by graph convolutional layers via initial residual identity mapping, resulting in informative high-order features for MSRA. A GraphNorm method is employed in the GNN to tackle over-smoothing problems and improve generalizability effectively. Then, based on real-time data, the risks of the multiple types of stability can be simultaneously and continuously predicted by the GNN, and the stable and unstable operation regions (SURs) can be visualized based on alpha shapes. The effectiveness of the proposed method is verified in the IEEE 39-bus system, the 179-bus western electricity coordinating council (WECC) system, and the Great Britain (GB) system. The comparison results of SURs associated with multi-stability are demonstrated and discussed to prioritize major types of stability problems.