TY - GEN
T1 - Integration of Machine Learning-Based CIE Standard Skies Model With Daylight Simulation for Building Energy Performance Analysis
AU - Aghimien, Emmanuel Imuetinyan
AU - Tsang, Ernest Kin Wai
AU - Li, Danny Hin Wa
AU - Wang, Zhenyu
N1 - Publisher Copyright:
© 2025, Free University of Bozen Bolzano. All rights reserved.
PY - 2025
Y1 - 2025
N2 - Daylight illuminance data is required for daylight scheme evaluations.while the adoption of daylight-linked control (DLLC) systems is a useful strategy for attaining energy savings on lighting. For evaluations of these daylighting schemes and DLLC, determining the sky conditions through sky luminance distributions is required. Moreover, through these distributions, the 15 CIE standard skies can be identified and daylight illuminance for any surface of interest can be derived. The crucial issue is that sky luminance data is sparingly measured. Recent studies have shown that the use of accessible climatic data and machine learning (ML) models for determining the standard skies can be viable alternatives. In this study, an ensembledbased Light Gradient Boosting Machine (LGBM) was used to identify the standard sky types in Hong Kong. The predictions of the LGBM model were then integrated with RADIANCE and EnergPlus for daylight and building energy simulations of a generic shopping mall. The simulation was carried out by comparing the Best fit, ASRC 1992 and the All-weather models against the measured data. Findings show that when tested, the LGBM model correctly classified the sky types over 70% of the time. Similarly, when used for daylight and energy simulations acceptable predictions were obtained from all models. Finally, it was found that the impact of the sky luminance distribution model on illuminance prediction is higher than that for energy estimation.
AB - Daylight illuminance data is required for daylight scheme evaluations.while the adoption of daylight-linked control (DLLC) systems is a useful strategy for attaining energy savings on lighting. For evaluations of these daylighting schemes and DLLC, determining the sky conditions through sky luminance distributions is required. Moreover, through these distributions, the 15 CIE standard skies can be identified and daylight illuminance for any surface of interest can be derived. The crucial issue is that sky luminance data is sparingly measured. Recent studies have shown that the use of accessible climatic data and machine learning (ML) models for determining the standard skies can be viable alternatives. In this study, an ensembledbased Light Gradient Boosting Machine (LGBM) was used to identify the standard sky types in Hong Kong. The predictions of the LGBM model were then integrated with RADIANCE and EnergPlus for daylight and building energy simulations of a generic shopping mall. The simulation was carried out by comparing the Best fit, ASRC 1992 and the All-weather models against the measured data. Findings show that when tested, the LGBM model correctly classified the sky types over 70% of the time. Similarly, when used for daylight and energy simulations acceptable predictions were obtained from all models. Finally, it was found that the impact of the sky luminance distribution model on illuminance prediction is higher than that for energy estimation.
UR - https://www.scopus.com/pages/publications/105014508981
U2 - 10.13124/9788860462022_25
DO - 10.13124/9788860462022_25
M3 - Conference contribution
AN - SCOPUS:105014508981
SN - 9788860462022
T3 - Building Simulation Applications
SP - 191
EP - 198
BT - Building Simulation Applications, BSA 2024 - 6th IBPSA-Italy Conference
A2 - Pernigotto, Giovanni
A2 - Patuzzi, Francesco
A2 - Gasparella, Andrea
A2 - Ballarini, Ilaria
A2 - Corrado, Vincenzo
A2 - Prada, Alessandro
T2 - 6th Building Simulation Applications conference, BSA 2024
Y2 - 26 June 2024 through 28 June 2024
ER -