Abstract: With the increasing penetration of renewable energy in active distribution network, how to effectively describe the uncertainty of the output of the renewable energy is a great challenge for active distribution network's planning and dispatching. For this problem, a multi-source-load scenario generation method of active distribution network based on variational autoencoder (VAE) is proposed. The label value is added to the traditional VAE unsupervised structure so that it has the ability of condition fitting, and an encoder and decoder network structure are designed using the graph neural network and the temporal convolutional network to extract the non-Euclidean and the temporal correlation characteristics respectively. This paper tests the proposed method with multi-source-load data, the results of which show that the proposed model can accurately restore the scenario, effectively describe the correlation between the multi-source-load scenarios, and the model can tag the data sample with unsupervised learning, then generate the same type of scenarios with the label value.