A comparison of six metamodeling techniques applied to building performance simulations

Research output: Contribution to journalJournal article

Abstract

Highlights
•Linear regression (OLS), support vector regression (SVR), regression splines (MARS).
•Random forest (RF), Gaussian processes (GPR), neural network (NN).
•Accuracy, time, interpretability, ease-of-use, model selection, and robustness.
•13 problems modelled for 9 training set sizes spanning from 32 to 8192 simulations.
•Methodology for comparison using exhaustive grid searches and sensitivity analysis.
Close

Details

Highlights
•Linear regression (OLS), support vector regression (SVR), regression splines (MARS).
•Random forest (RF), Gaussian processes (GPR), neural network (NN).
•Accuracy, time, interpretability, ease-of-use, model selection, and robustness.
•13 problems modelled for 9 training set sizes spanning from 32 to 8192 simulations.
•Methodology for comparison using exhaustive grid searches and sensitivity analysis.
Original languageEnglish
JournalApplied Energy
Volume211
Pages (from-to)89-103
ISSN0306-2619
DOI
StateE-pub ahead of print - 2018
Publication categoryResearch
Peer-reviewedYes

    Research areas

  • Gaussian process regression (kriging), Random forest, Neural network, Support vector regression, Sensitivity analysis, Supervised learning
ID: 265086489