Forced oscillation location based on temporal graph convolutional network

Oscillation source location is the key to suppressing the forced oscillation and guaranteeing power system stability. For forced oscillation events with clear disturbance sources, locating the source of the oscillation quickly and accurately is the primary task to achieve forced oscillation suppression. The existing methods cannot make full use of the spatial topology information of the power grid and the temporal characteristics of oscillations, which limits the location accuracy. Therefore, a forced oscillation location method based on temporal graph convolution network is proposed. Firstly, the graph data is constructed according to node features and topology information of the power grid, and the oscillation space features are extracted by graph convolution neural network. At the same time, the temporal correlation of oscillation data of multiple nodes is extracted by the gated recurrent unit neural network. The spatial and temporal characteristics are fused by the spatiotemporal graph convolution unit. Then, the forced oscillation location is modeled as a classification problem, and the location model based on temporal graph convolution neural network is trained. The case analysis shows that the method proposed has higher accuracy and better generalization. The method proposed still has good performance in the case of data noise and missing.