C. Gkoutroumpi, N. Virtsionis Gkalinikis, D. Vrakas. SGAN: Appliance signatures data generation for NILM applications using GANs, To be presented in the 12th Computing Conference 2024, London UK, (2024)
Author(s): C. Gkoutroumpi, N. Virtsionis Gkalinikis, D. Vrakas
Appeared In: To be presented in the 12th Computing Conference 2024, London UK
Keywords: Synthetic Data, Deep Learning, Energy, Timeseries, GAN, NILM, Generative AI
Tags:
Abstract: The development and evolution of advanced energy system technologies is one of the most important goals for the global commu- nity in recent years. In this effort, the utilization and analysis of energy time series is of decisive importance for the understanding of energy con- sumption and production patterns. However, access to real data may be limited due to the sensitivity of the information and the limited amount of data already available. This has led to the use of methods to produce artificial data in order to enrich existing datasets. Generative Adversarial Networks or GANs are an approach to generative modeling using deep learning methods based on the logic of adversarial learning, and consist of two adversarial neural networks, a generator and a discriminator, which work together to produce realistic and unbiased data. The subject of the current paper is the creation of a GAN pipeline capable of producing power time series that resemble those observed in the real world, pre- serving the main characteristics and diversity of the observed electrical devices. The proposed method shows promising results, outperforming other state-of-the-art models in two calculated metrics.