Research on power system dynamic frequency prediction method based on Theory-Informed graph convolution and Informer

Since the massive integration of renewable generation and the complex interconnection of large regional power grids, the dynamic behavior of the system frequency after disturbance is more and more complex, and the traditional method has disadvantages such as poor accuracy, high data demand and insufficient migration ability. In this paper, a method of power grid dynamic frequency prediction based on Physics-Informed graph convolution and Informer is proposed. Firstly, graph convolution and Informer model are used to deal with the spatial and temporal distribution of frequency response variables in the system. Secondly, according to the characteristics of the model and the frequency response characteristics of the system, the message passing, input characteristics and loss function of the proposed model are constructed by theoretically guided deep learning method, so as to realize the fusion of theory and data. Then, considering the uncertainty factors such as random fluctuations of the system source load and real-time changes of parameters, combined with the system parameters obtained by wide-area measurement of PMU, a power grid dynamic prediction process based on the theory guided graph convolution and Informer model is proposed to achieve model migration. Finally, IEEE39 nodes and a provincial power grid are used to verify the effectiveness of the proposed method