Research on Power Load Data Enhancement Algorithm Based on Metric Learning and Variational Autoencoder
In the field of power time series data,the training of traditional models on such datasets faces significant challenges due to issues such as privacy protection,small data size and imbalance.To this end,this paper proposes a novel time series data generation framework that combines Variational Autoencoder(VAE)with metric learning.The method constructs discriminative VAE potential space through metric learning to enhance the effectiveness of data augmentation;meanwhile,it utilizes a time series trend decomposition algorithm to inject the original data structure to enhance the interpretability of the generated data.The experimental results show that compared with the existing SOTA methods,the algorithm excels in both data generation quality and model performance,with an average speedup of 37.45%,significantly reduces the cost of data collection and labeling,and has the potential for cross-domain time series data applications.
electricity time series datavariational autoencodermetric learningtime series decomposition
万方数据
