Transient stability analysis and emergency generator tripping control of power system based on RNN and GCNII

To achieve rapid and precise determination of post-fault system stability and to facilitate effective corrective actions, this work introduces a methodology built upon deep learning architectures that process graph-structured data with both temporal and spatial dimensions. The proposed framework fulfills two main functions: it classifies the stability status of the power grid after contingencies, and it predicts which generators tend to oscillate coherently. Based on these predicted coherent groups and an equivalent single-machine infinite-bus model, a generator tripping strategy is subsequently devised. The procedure begins with the construction of feature vectors derived from energy functions associated with transient phenomena, followed by the formation of graph samples that mirror the physical connectivity of the network. Next, a recurrent neural network is utilized to capture the temporal evolution of the features, while a graph convolutional network featuring residual connections and identity mapping is applied to extract spatial dependencies. The combined outputs enable both stability classification and coherent grouping. Finally, the tripping scheme is established using the grouping results and the equivalent model. This multi-stage prediction framework seamlessly unites stability evaluation with emergency control actions. The performance of the proposed method is verified through case studies on a standard test system.