Artificial neural network intelligent technique and multiple nonlinear regression for prediction and optimization of the transmittance of lightpipes and implementation in BIM

Daylighting features prominently in sustainable building design. It has been proven that daylighting not only saves the electric lighting energy consumption, but also improves the visual comfort and occupants’ health. A number of daylighting designs and control strategies have been presented and practised. Performance prediction of these designs is essential in daylighting research. The innovation of natural daylighting light pipe took place more than thirty years ago. However, no efficient and accurate prediction method, which includes the efficiency of straight light pipe, especially the bended light pipe has been made available. Therefore, a prediction model for light pipes is desirable to assess and predict its efficiency and potential in energy saving. This thesis attempts to develop an Artificial Neural Networks (ANNs) based prediction model for the performance of lightpipes and implement it in the Building Information Modelling (BIM) platform to help the designers, engineers and asset managers make informed decisions in daylighting lightpipes design. A comprehensive and critical literature review is first introduced covering the advanced artificial neural network intelligent technique in the application of the luminance and illuminance prediction, energy saving, daylighting controls and the optical property of lightpipes. An optical analysis software Photopia is employed to simulate the daylighting performance of light pipes to generate the real database and calculate the efficiency of the light pipes. It is then followed by ANNs simulations in Matlab for forming a forecasting model for light pipe performance. To empower the prediction model and make it easy and friendly to be used, the developed ANNs model for lightpipe performance is innovatively implemented in BIM software Revit, as a plug-in application tool. This tool in Revit enables the prediction of the transmittance of lightpipes directly without running the programme in Matlab. It can help the designers or users choose the lighpipe parameters easily and accurately and therefore add value to the industry and the research community