Decomposition-based Data Augmentation for Time-series Building Load Data
Yang Deng, Rui Shi Liang, Dan Wang, Ao Li, Fu Xiao · 2023
Building load data, i.e., building electricity demands, are important for many downstream applications such as load forecasting, demand response, and others. Recent applications, in particularly, those based on machine learning models, require a large amounts of data. Unfortunately, many buildings do not have sufficient data. To augment data, recent schemes are relying on generative adversarial networks (GANs). GAN-based schemes can generate new samples for the same distribution, i.e., to enrich data diversity. However, they are not suitable for augmenting the data with insufficient data distributions, e.g., a data shortage caused by insufficient time coverage, a common problem for new buildings. This paper aims to address this problem.