Development of a Machine-Learning Framework for Overall Daylight and Visual Comfort Assessment in Early Design Stages
Application of machine learning methods as an alternative for building simulation software has been progressive in recent years. This research is mainly focused on the assessment of machine learning algorithms in prediction of daylight and visual comfort metrics in the early design stages and provid...
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
SolarLits
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/be97f02be2704d5aa130154e4b7a94ee |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Application of machine learning methods as an alternative for building simulation software has been progressive in recent years. This research is mainly focused on the assessment of machine learning algorithms in prediction of daylight and visual comfort metrics in the early design stages and providing a framework for the required analyses. A dataset was primarily derived from 2880 simulations developed from Honeybee for Grasshopper. The simulations were conducted for a side-lit shoebox model. The alternatives emerged from different physical features, including room dimensions, interior surfaces’ reflectance factor, window dimensions, room orientations, number of windows, and shading states. Five metrics were applied for daylight evaluations, including useful daylight illuminance, spatial daylight autonomy, mean daylight autonomy, annual sunlit exposure, and spatial visual discomfort. Moreover, view quality was analyzed via a grasshopper-based algorithm, developed from the LEED v4 evaluation framework. The dataset was further analyzed with an artificial neural network algorithm. The proposed predictive model had an architecture with a single hidden layer consisting of 40 neurons. The predictive model learns through a trial and error method with the aid of loss functions of mean absolute error and mean square error. The model was further analyzed with a new set of data for the validation process. The accuracy of the predictions was estimated at 97% on average. The View range metric in the quality view assessment, mean daylight autonomy and useful daylight illuminance had the best prediction accuracy among others respectively. The developed model which is presented as a framework could be used in early design stage analyses without the requirement of time-consuming simulations. |
|---|