Comparision of Different Data-Driven Models on Prediction of Useful Daylight Illuminance (UDI)

EasyChair Preprint 6389

Abstract

Energy consumption of the built environment worldwide is increasing at a faster rate than the population. With building energy consumption, the amount of greenhouse gas released into the atmosphere also increases, further enhancing global warming. In this context, lighting is seen as one of the most critical issues associated with energy consumption. Despite many studies related to the evaluation of daylighting in specific buildings, few studies examine daylighting in the urban context. Therefore, daylighting in dense urban areas should be analyzed and predicted for optimization. The predictions using surrogate models based on machine learning approaches have the potential to predict daylighting results and reduce the environmental impact of the buildings. This research aims to enhance the applications of machine learning approaches in daylighting predictions in the urban context. As a case study, a small urban area in Turkey, Ankara, is simulated, and different machine learning models (i.e., Multiple Linear Regression, Artificial Neural Network, Random Forest) are presented. Performances of prediction models will be compared.

Keyphrases: Daylighting, daylighting metrics, machine learning, prediction models

@booklet{EasyChair:6389,
  author    = {İlkim Canlı and Orçun Koral İşeri and İpek Gürsel Dino},
  title     = {Comparision of Different Data-Driven Models on Prediction of Useful Daylight Illuminance (UDI)},
  howpublished = {EasyChair Preprint 6389},
  year      = {EasyChair, 2021}}