基于深层级联残差图卷积的暂态稳定评估模型及其实际电网应用OA北大核心CSTPCD

The studies and tests of the data-driven transient power angle stability assessment models are mainly carried out in small-scale example systems,with insufficient application tests in real power grids.The internal reason is that the contradiction between the local information extraction characteristics and the global power angle stability in graph deep learning has not been solved.In this paper,a deep cascade residual map convolution model is designed to achieve an effective improvement in model layer stacking performance through a deep cascade structure containing residual connectivity.It also proposes the MinMaxPooling module to decouple the model parameters from the system size.The model structure design independent of the number of nodes can solve the problem of data-driven models applied to large-scale real power grids.The validity and performance of the proposed model are tested on a real regional power grid with 5 419 nodes.