Topological Convolutional Neural Networks for Transient Stability Assessment on Massive Historical Online Power Grid Data
Xinglei Chen, Yanhao Huang, Songtao Zhang, Wenchen Li, Shujun Zhang · 2020
The power system is a highly nonlinear complex dynamic system with chaos. Based on the historical online data generated by dynamic stability analysis (DSA) system, transient stability assessment (TSA) approach enables rapid screening of transient stability fault sets, and thus improving the operation stability of the power grid. We propose a novel deep learning framework, called Topological Convolutional Networks (TCNN), to specifically model the topological structures of power grids and efficiently extract features. Based on the adjacent matrix, topological convolution updates the node’s representation using the representation of its neighbors, and topological pooling operation summarizes the topological local features. The results show that the proposed model outperforms other strong baselines by a large margin on both measurements of compressing the candidate fault set and reducing the misclassification of unstable samples.