Optimization of Complex Fenestration Systems using an Artificial Neural: Network Considering Energy and Daylighting Performance of Office Buildings
Authors: Uribe, Daniel; Veraand, Sergio; Bustamante, Waldo
The use of artificial neural networks (ANNs) in building performance problems has been widely studied by different authors in the last years. ANNs can decrease the computational time when the building design is complex due to high number of variables. In this research, an ANN was developed in Python and used to optimize an office space with exterior and fixed complex fenestration systems and dimmed luminaries in three different climates of Chile considering variables such as window‐to‐wall ratio, solar heat gain coefficient, U‐value of windows, shading device, walls’ thermal resistance and insulation position. The office performance metrics considered in the objective function of optimization process are total energy consumption (sum of lighting, heating and cooling energy consumption) and two visual comfort criteria, spatial daylight autonomy (sDA) and annual sunlight exposure (ASE). A total of 5,400 lighting simulations and 12,800 energy simulations were performed to train the ANN. The simulations were carried out using mkSchedule, a tool that integrates energy and lighting simulations. The results show the capability of the ANN to be incorporated to an optimization process of office buildings based on energy performance and visual comfort metrics.