Affiliations: [a]
College of Environment and Civil Engineering, Chengdu University of Technology, Chengdu, Sichuan, P. R. China | [b]
Business School, Sichuan University, Chengdu, Sichuan, P. R. China | [c]
School of Architecture, Southwest Jiaotong University, Chengdu, Sichuan, P. R. China | [d]
College of Geophysics, Chengdu University of Technology, Chengdu, Sichuan, P. R. China | [e]
School of Management, Zhejiang University of Technology, Hangzhou, Zhejiang, P.R. China
Abstract: The issue of how to fully utilize natural daylighting of public buildings is one of the greatest practical objectives for lighting savings. The rapid and accurate prediction of the daylighting coefficient at the early design stage can provide a quantitative basis for energy-saving optimization. However, it is not comprehensive to determine the design parameters according to experience. The key problem that is still facing designers is the interoperability between building modeling and energy simulation tools. In this paper, an integrated approach using a dataset created by building information modeling and artificial neural network technology is developed for the fast optimal daylight factor prediction of large public spaces at the early design stage. According to this approach, the value of daylight factors is calculated for different windowsill heights, window heights and widths by Autodesk® Revit and Ecotect Analysis to form a dataset. With this dataset, an artificial neural network model is established using the backpropagation algorithm to predict the relevant design parameters. With their large interior spaces, the reading areas of the aboveground five floors in Chengdu University of Technology Library are selected to carry out the daylight factor experiment and rapid prediction. A total of 495 groups of experimental data are randomly divided into training and testing sets. The root mean squared errors are below 0.1, which indicates a high regression model fitting. A total of 225,369 groups of prepared data are used in the prediction model to obtain the optimal windowsill height (1.0 m), window height (2.4 m) and window width (2.1 m) for five floors in the case of the maximum daylighting coefficient. Finally, a smartphone app is designed to facilitate daylight factor prediction without any experience in modeling and simulation tools, which is simple and available to realize prediction visualization and historical result analysis.
Keywords: Daylight factor, rapid prediction, building information modelling, artificial neural network, library, app
