A dimensionality reduction method to select the most representative daylight illuminance distributions

Abstract

ABSTRACTOne challenge when evaluating daylight distribution is dealing with the large amount of temporal and spatial data, visualizations and variability in illuminances that are assessed in buildings. Using a dimensionality reduction method based on principal component analysis, we identified the most representative annual daylight distributions. We modelled a rectangular room containing an analysis grid of 3200 illuminance sensor points and simulated 3285 different temporal daylight conditions using an annual occupancy schedule ranging from 08:00 to 17:00 with one-hour sampling intervals in two locations: Singapore and Oakland, California. Our approach explained 98% of the illuminance variability with three daylight distributions in Singapore, and 92% using six in Oakland, California. Our dimensionality reduction strategy was also generalized using a complex building geometry showing the utility of the method. We think this approach can be used to provide a more efficient and reliable method to analyse daylight performance in building practice.