Generative Adversarial Network Applied to the Energy Efficiency of Buildings

Abstract

Energy consumption in buildings represents a significant proportion of global energy consumption, which raises the need to develop strategies for its optimization. However, datasets can often be incomplete when analyzing energy variables such as electricity consumption due to missing measurements or equipment failures. Generative antagonistic networks (GANs) can generate high-quality synthetic data that mimic actual data distribution. Through a literature review, this paper examines how GANs have been applied to study building energy efficiency. In addition, as a case study, it considers a dataset generation from historical data of the FCMF-UG building of the University of Guayaquil. The findings demonstrated in the case study that variability of the original data influences the results of curve generation with GANs. These preliminary results can be a baseline for future analysis of GANs applied to building energy efficiency.