Paper Conference

2018 Building Performance Analysis Conference and SimBuild co-organized by ASHRAE and IBPSA-USA

     

Computing Long-Term Daylighting Simulations from High Dynamic Range Imagery Using Deep Neural Networks

Yue Liu, Alex Colburn, Mehlika Inanici
University of Washington, Seattle, WA
Zillow Group, Seattle, WA

Abstract: Compared with illuminance-based metrics, luminance-based metrics and evaluations provide better understandings of occupant visual experience. However, it is computationally expensive and time consuming to incorporate luminance-based metrics into architectural design practice because annual simulations require generating a luminance map at each time step of the entire year. This paper describes the development of a novel prediction model to generate annual luminance maps of indoor space from a subset of images by using deep neural networks (DNNs). The results show that by only rendering 5% of annual luminance maps, the proposed DNNs model can predict the rest with comparable accuracy that closely matches those high-quality point-in-time renderings generated by Radiance (RPICT) software. This model can be applied to accelerate annual luminance-based simulations and lays the groundwork for generating annual luminance maps utilizing High Dynamic Range (HDR) captures of existing environments.
Pages: 119 - 126
Paper:
simbuild2018_C018