Daylight factor estimation at an early design stage to reduce buildings' energy consumption due to artificial lighting: A numerical approach based on Doehlert and Box–Behnken designs

• Regression models for daylight factor forecasting in early design stage.
• 2 Case studies with different characteristics.
• Simulation scenarios based on Doehlert and Box–Behnken design.
• Development of polynomial regression models.
• Validation and evaluation of polynomial regression models.

In buildings of residential and tertiary sectors, daylight issues primarily respond to increasing needs regarding the occupants' well being and visual confort as well as to the supply of natural light for working, by decreasing energy consumption due to necessary artificial lightning for the internal human activities. Daylight optimization at an early design stage can be referred to a way of designing architectural forms that take advantage of the prevailing urban context, such as existing shadow masks, to achieve a comfortable interior environment while minimizing energy use and reliance on artificial lighting systems. Our past study concluded in the development of EcCoGen, which is a kind of software that belongs to the family of tools based on interactive generative genetic algorithm optimization. The software generates solutions evaluated according to certain criteria, including mainly energy performance issues. Since the energy evaluation is included in the software features, we aimed to enhance software's applicability including energy reduction due to lighting issues via visual comfort optimization. Based on this objective, in this paper we describe a simple methodology to optimize the daylight potential of an architectural form at an early design stage. A numerical approach conducted employing the DIALux 4.5© is presented while a variety of simulation scenarios has been investigated on the basis of Doehlert and Box–Behnken DOE (design of experiments) methods. Our purpose is to develop accurate enough regression models for daylight factor prediction at an early design stage, when the problem's data are not precisely determined (dimension of glazing area, materials, opacities). To validate this statistical forecast system, many simulation scenarios were carried out and the statistical results are in compliance with the numerical simulations. The regression models' results show that the error caused by simplification is acceptable in most conditions, and a lot of coupling calculation is saved. Finally, a formal error analysis for the resulted regression models has been conducted to validate its forecast capacity. As a conclusion the statistical reduction of complex numerical modeling to simple regression models in the form of polynomial equations aims to assist architects and engineers to directly obtain a high precision estimation of their architectural form's daylight potential at an early design stage.