Source-Load-Topo Joint Scenario Generation Based on Graph Variational Auto-Encoder

Scenario generation is a key approach for analyzing uncertainty in power systems. However, existing methods focus solely on generating source-load node states while overlooking topology uncertainty, which limits the accuracy and comprehensiveness of the generated scenarios. To bridge this gap, a source-load-topo joint scenario generation model, SLT-VAE, is proposed to simultaneously determine the corresponding topology states when generating the source-load scenarios. SLT-VAE comprises a spatial-temporal graph encoder and a joint scenario graph decoder. Specifically, the encoder utilizes spatial-temporal attention to adaptively extract information across spatial-temporal dimension. A spatial-temporal convolution is applied to fuse the spatial dependency among source-load nodes and temporal features in different time steps to obtain a dense latent variable containing multi-scale spatial-temporal information. The decoder then outputs the source-load node attribute matrix and its corresponding adjacency matrix in a one-shot manner, leveraging this variable to simultaneously generate source-load and topology scenarios, while quantifying the uncertainty in the topology state. Furthermore, the simulated grid dataset containing 100 different topology scenarios is provided to validate the effectiveness of the proposed method. Experimental results indicate that SLT-VAE can capture complex spatial-temporal correlations and dynamics. Its generated source-load scenarios with topology states is as real as ground-truth data. This facilitates a more comprehensive source-load scenario generation, thus enhancing the operation and planning of power system.