A source-load scene countermeasure generation method based on a relational graph convolution network

Abstract

The application discloses a source-load scene adversarial generation method based on a relational graph convolution network, and comprises the following steps: step 1, source-load scene data is acquired to construct a graph structure representation of a real source-load scene; step 2, a graph generative adversarial network integrating a relational graph convolution network is established, the graph generative adversarial network model comprises a source-load scene data generator based on a multilayer perceptron and a source-load scene data discriminator based on the relational graph convolution network, and the graph generative adversarial network is trained; and step 3, random noise is regenerated and input to the source-load scene data generator in the trained graph generative adversarial network, so that a current source-load scene graph structure representation is obtained and used as a source-load scene simulation sample. The application can enhance the ability to cope with random fluctuations and topological changes of source-load power, and guarantee the safe and economic operation of a power grid.