Prediction of indoor illuminance using one external sensor

The thesis ”Predicting Indoor Illumination Using One External Sensor” by Hans-Andrea Danuser explores the usage of machine learning algorithms to predict indoor illumination based on external environmental data. The thesis was conducted as continuation of a series of studies made by the Human-IST Research Institute in Fribourg and IDIAP in Martigny, focusing on the interplay between energy reduction and occupant comfort in workplace illumination. The primary goal was to assess visual and thermal comfort as well as energy savings through an integrated control scheme using daylight and glare predictions. The research involved creating a virtual 3D model called RADIANCE to predict indoor light values and calibrating this model with real-world data from various office rooms. The thesis is trying to omit the RADAINCE model and directly predict the indoor illumination. Several popular Machine learning models, particularly Gradient Boosting Decision Trees (GBDT) and Neural Networks (NN) were evaluated for their accuracy in predicting indoor illumination levels. These models used data including time, external illuminance and sun position. Key findings include the successful application of machine learning algorithms in predicting indoor lighting conditions. The study also highlights the challenges and limitations faced due to the quality of available data and the need for comprehensive, long-term datasets for more accurate predictions. The research concludes with the potential of these algorithms for practical applications in predicting and managing indoor lighting environments.

Supervisors: Moreno Colombo, Julien Nembrini

Student: Hans-Andrea Danuser

Project status: Finished

Year: 2023

Document: Report