Paper Conference

ASHRAE & IBPSA-USA SimBuild 2016: Building Performance Modeling Conference


Comparison of Regression Techiques for Surrogate Models of Building Energy Performance

Arfa Aijazi, Leon Glicksman
Massachusetts Institute of Technology, Cambridge, MA

Abstract: Surrogate modeling is a technique to approximate the behavior of complex systems based on a limited set of computationally expensive simulations. Such models can permit real time optimization of building designs. There are many ways to develop the surrogate model however, the relative advantages and disadvantages of one modeling technique versus another are not well understand, so modelers lack a criteria for selecting amongst approaches. The objective of this paper is to compare the accuracy and computation time of multiple methods. This paper also explores the choice of design variables and parameter combinations to estimate the building energy consumption when climate is one of the variables. The results of this analysis will contribute to selecting a regression technique that limits errors relative to a detailed simulation.
Pages: 327 - 334