Power System Transient Stability Assessment Method Based on Graph Convolutional Network Considering Unbalanced Samples

Purposes Data-driven transient stability assessment of power systems has become the mainstream research direction at this stage. However, in the practical applications, there are still problems such as too few unstable samples and lack of consideration of the impact of power system spatial topology information on transient stability assessment. In view of these issues, a new transient stability assessment model based on conditional generative adversarial network (CGAN) and graph convolution network (GCN) is proposed. Methods First, CGAN was used to perform targeted enhancement on sparsely distributed unstable samples as a link between the original unbalanced data set and the data-driven transient stability discrimination method, so as to achieve accurate optimization of the extremely unbalanced original data set. Then, the spatial topology information of the power grid was introduced as input, and the GCN was used to mine the spatial feature relationship of the power grid. After that, the transient stability assessment model was constructed by combining the feature vector of the node itself and its transient stability label, which enhances the model’s generalization ability for changes in power grid operation mode and topological structure. Finally, simulation verifications were carried out on the IEEE-39 node system and the IEEE-118 node system. Conclusions The results show that the proposed CGAN-GCN transient stability assessment model has improved accuracy and demonstrates strong generalization ability during model topology changes.