A Clustering-Generative Model Based Method for Load Data Augmentation

As big data technologies become more prevalent in the energy sector, the importance of data is increasing. Data augmentation techniques can enhance the size and quality of data sets. In the scenario of an integrated energy system, the complex coupling relationship of various forms of energy poses a challenge for load data augmentation, for which a data augmentation method for electricity and thermal coupled load is proposed in this paper. First, an asymmetric Variational Autoencoder (VAE) with KL cost annealing is trained. The encoder part is used as a representation learner to extract the electricity and thermal features, based on which K-means++ is used to cluster the raw data. Then the decoder part generates new samples proportionally according to the clustering results. The experimental results show that the load data generated by this method can retain the overall distribution characteristics and the coupling relationship between electricity and thermal.