Automatic Completion Method of Power Missing Data in Distribution Network Based on GCN-VAE Model

With the expansion of distribution network and extensive information collection, some data are missing due to interference in data collection and transmission, so it is urgent to realize effective data completion. This paper presents a novel approach for completing missing power data using a combination of graph convolution neural network (GCN) and variational self-encoder (VAE). Firstly, the power data is passed through GCN, and the spatial characteristics are learned from the original data, so as to obtain the spatial dependence. Then, the data is reconstructed through VAE, and the network parameters are constantly updated by calculating the loss function value during training. Using the two-way algorithm model of GCN-VAE, the forward completion data and reverse completion data of time series are weighted, and the missing data is automatically completed. The experimental results indicate that the GCN-VAE-based completion model achieves higher accuracy compared to traditional data completion methods.