상세정보
Export to Refworks부가기능
Computing Long-Term Daylighting Simulations from High Dynamic Range Imagery Using Deep Neural Networks
상세 프로파일
| 자료유형 | 학위논문 |
|---|---|
| 서명/저자사항 | Computing Long-Term Daylighting Simulations from High Dynamic Range Imagery Using Deep Neural Networks. |
| 개인저자 | Liu, Yue . |
| 단체저자명 | University of Washington. Built Environment. |
| 발행사항 | [S.l.]: University of Washington., 2019. |
| 발행사항 | Ann Arbor: ProQuest Dissertations & Theses, 2019. |
| 형태사항 | 177 p. |
| 기본자료 저록 | Dissertations Abstracts International 81-03B. Dissertation Abstract International |
| ISBN | 9781085730839 |
| 학위논문주기 | Thesis (Ph.D.)--University of Washington, 2019. |
| 일반주기 |
Source: Dissertations Abstracts International, Volume: 81-03, Section: B.
Advisor: Inanici, Mehlika. |
| 이용제한사항 | This item must not be sold to any third party vendors.This item must not be added to any third party search indexes. |
| 요약 | Practitioners and researchers utilize long-term daylighting analysis to evaluate the luminous environment under a wide range of naturally occurring sky and sun conditions. Annual simulation tools and metrics are commonly based on illuminance calculations, which focus on daylight availability and potential energy savings. Design and research practices are shifting from illuminance-based metrics towards luminance-based (human-centric) metrics and evaluations. However, hourly annual luminance maps require either labor-intensive and computationally expensive simulation processes or impracticable long-term in-situ measurements. Simulations are performed by rendering the environment using physically-based rendering techniques. Annual luminance-maps are generated by solving the radiance regression function, which is a non-linear mapping of pixel-scale luminance values from local and contextual attributes of surface points for every hour of daylight (approximately 4,000 simulations). This process can take weeks to compute with complex building geometries.This dissertation presents a novel data-driven machine learning approach to make annual luminance-based evaluations more efficient and accessible. The methodology is based on predicting annual luminance maps from a limited number of point-in-time high dynamic range (HDR) imagery by utilizing a deep neural network (DNN).An initial study used a small dataset to train a DNN model that accurately predicts annual luminance maps. The training dataset utilizes 200 hourly samples from the annual luminance map dataset, approximately 5% of the entire dataset. A second study refines the methodology of the first study and applies it to panoramic luminance maps (with 360째 horizontal and 180째 vertical field of view). Panoramic views are more data-intensive compared to perspective views that have a much narrower field of view, but they provide better information about an occupant's visual experience. A thorough sensitivity analysis is performed to develop guidelines for determining the minimum and optimum data collection periods, and the minimum number of imagery needed for accurately generating the annual luminance maps. The analyses suggest that the proposed DNN model can faithfully predict high-quality panoramic luminance maps from one-month hourly imagery generated or collected during daylight hours around the equinoxes. Alternatively, 9 days of hourly data collected around the spring equinox, summer solstice, and winter solstice would suffice (2.5% of total annual renderings) to predict the point-in-time luminance maps for the rest of the year. Data generation is greatly accelerated with this method, and only requires 30 minutes of training time on a NVIDIA GeForce 1080Ti GPU and a few seconds of computing time. Therefore, this approach alleviates the problem of extensive calculation time that typically hampers the utilization of annual luminance-based daylighting simulations and metrics.With the developed workflow, practitioners and researchers can efficiently incorporate long-term luminance-based metrics into the design and research process. This study also lays the groundwork for generating annual luminance maps utilizing short-term HDR photographs of existing environments, which will enable quantitative analysis of daylit environments without requiring a time-consuming modeling process. |
| 일반주제명 | Architecture. Artificial intelligence. |
| 언어 | 영어 |
| 바로가기 | 
: 이 자료의 원문은 한국교육학술정보원에서 제공합니다. |
