Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Machine learning-based framework to predict single and multiple daylighting simulation outputs using neural networks

Rania Labib
Prairie View A&M university, United States of America

Abstract: Building energy consumption accounts for 30% of global energy consumption (EIA, 2017). To support the development of energy-efficient built environments and cities, architects, urban planners, and engineers have begun to utilize building performance simulation (BPS). Supporting decision-making and steering the design toward high performance is crucial in the early design phase when decisions have the biggest impact on the final building’s energy consumption and costs (Attia et al., 2012; Hygh et al., 2012; Kanters & Horvat, 2012). However, BPS tasks are usually time-consuming. Therefore, there is a need for a framework that would speed up the BPS process. This paper aims to develop a machine learning (ML) algorithm, specifically neural networks (NN), that can potentially speed up the process of daylighting simulations by executing only a small subset of the simulations to predict the performance of daylighting of thousands of design configurations. Furthermore, the paper will investigate the use of NN to predict single and multiple outputs of point-in-time and annual based daylighting simulations respectively.
Keywords: Machine Learning, Neural Network, Daylighting
Pages: 1334 - 1340