Skip to main navigation Skip to search Skip to main content

Comparative Analysis of Machine Learning Models Optimized by Bayesian Algorithm for Indoor Daylight Distribution Prediction

  • Unhai Shen
  • , Yunsong Han*
  • *Corresponding author for this work
  • Harbin institute of technology
  • Ministry of Industry and Information Technology

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Daylight distribution evaluation is vital for daylight design. However, its application in the early design stage is limited due to the time-consuming simulation process. Many statistical models were proposed to reduce the prediction time, yet application of the machine learning model in daylight prediction was relatively rare and has very limited generalization capability. This paper aims to propose a new workflow for indoor daylight distribution prediction, and compare the performance of XGB, RF, SVR and MLP models with Bayesian optimization. The results showed the MLP based prediction model achieved best generalization performance for indoor daylight prediction, which reduced the simulation time to less than 1 second and maintained satisfactory accuracy.

Original languageEnglish
Title of host publicationPLEA 2020 - 35th PLEA Conference on Passive and Low Energy Architecture Planning Post Carbon Cities, Proceedings
EditorsJorge Rodriguez Alvarez, Joana Carla Soares Goncalves, Joana Carla Soares Goncalves, Joana Carla Soares Goncalves
PublisherUniversity of A Coruna and Asoc
Pages988-993
Number of pages6
ISBN (Electronic)9788497497947
StatePublished - 2020
Externally publishedYes
Event35th PLEA Conference on Passive and Low Energy Architecture Planning Post Carbon Cities, PLEA 2020 - A Coruna, Spain
Duration: 1 Sep 20203 Sep 2020

Publication series

NamePLEA 2020 - 35th PLEA Conference on Passive and Low Energy Architecture Planning Post Carbon Cities, Proceedings
Volume2

Conference

Conference35th PLEA Conference on Passive and Low Energy Architecture Planning Post Carbon Cities, PLEA 2020
Country/TerritorySpain
CityA Coruna
Period1/09/203/09/20

Keywords

  • Artificial Neural Network
  • Bayesian Optimization
  • Daylight Distribution Prediction
  • Machine Learning
  • UDI

Fingerprint

Dive into the research topics of 'Comparative Analysis of Machine Learning Models Optimized by Bayesian Algorithm for Indoor Daylight Distribution Prediction'. Together they form a unique fingerprint.

Cite this