Development of surrogate models using artificial neural network for building shell energy labelling

A.P. Melo, D. Cóstola, R. Lamberts, J. L M Hensen

Research output: Contribution to journalArticle

30 Citations (Scopus)

Abstract

Surrogate models are an important part of building energy labelling programs, but these models still present low accuracy, particularly in cooling-dominated climates. The objective of this study was to evaluate the feasibility of using an artificial neural network (ANN) to improve the accuracy of surrogate models for labelling purposes. An ANN was applied to model the building stock of a city in Brazil, based on the results of extensive simulations using the high-resolution building energy simulation program EnergyPlus. Sensitivity and uncertainty analyses were carried out to evaluate the behaviour of the ANN model, and the variations in the best and worst performance for several typologies were analysed in relation to variations in the input parameters and building characteristics. The results obtained indicate that an ANN can represent the interaction between input and output data for a vast and diverse building stock. Sensitivity analysis showed that no single input parameter can be identified as the main factor responsible for the building energy performance. The uncertainty associated with several parameters plays a major role in assessing building energy performance, together with the facade area and the shell-to-floor ratio. The results of this study may have a profound impact as ANNs could be applied in the future to define regulations in many countries, with positive effects on optimizing the energy consumption.

LanguageEnglish
Pages457-466
Number of pages10
JournalEnergy Policy
Volume69
DOIs
Publication statusPublished - Jun 2014

Fingerprint

artificial neural network
Labeling
shell
Neural networks
energy
Facades
typology
Sensitivity analysis
simulation
sensitivity analysis
Energy utilization
labelling
Cooling
cooling
climate
parameter
programme
Uncertainty

Keywords

  • artificial neural network
  • building energy simulation
  • surrogate model

Cite this

@article{1b2165418f7c4016911e88ab52c3ccba,
title = "Development of surrogate models using artificial neural network for building shell energy labelling",
abstract = "Surrogate models are an important part of building energy labelling programs, but these models still present low accuracy, particularly in cooling-dominated climates. The objective of this study was to evaluate the feasibility of using an artificial neural network (ANN) to improve the accuracy of surrogate models for labelling purposes. An ANN was applied to model the building stock of a city in Brazil, based on the results of extensive simulations using the high-resolution building energy simulation program EnergyPlus. Sensitivity and uncertainty analyses were carried out to evaluate the behaviour of the ANN model, and the variations in the best and worst performance for several typologies were analysed in relation to variations in the input parameters and building characteristics. The results obtained indicate that an ANN can represent the interaction between input and output data for a vast and diverse building stock. Sensitivity analysis showed that no single input parameter can be identified as the main factor responsible for the building energy performance. The uncertainty associated with several parameters plays a major role in assessing building energy performance, together with the facade area and the shell-to-floor ratio. The results of this study may have a profound impact as ANNs could be applied in the future to define regulations in many countries, with positive effects on optimizing the energy consumption.",
keywords = "artificial neural network, building energy simulation, surrogate model",
author = "A.P. Melo and D. C{\'o}stola and R. Lamberts and Hensen, {J. L M}",
year = "2014",
month = "6",
doi = "10.1016/j.enpol.2014.02.001",
language = "English",
volume = "69",
pages = "457--466",
journal = "Energy Policy",
issn = "0301-4215",

}

Development of surrogate models using artificial neural network for building shell energy labelling. / Melo, A.P.; Cóstola, D.; Lamberts, R.; Hensen, J. L M.

In: Energy Policy, Vol. 69, 06.2014, p. 457-466.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Development of surrogate models using artificial neural network for building shell energy labelling

AU - Melo, A.P.

AU - Cóstola, D.

AU - Lamberts, R.

AU - Hensen, J. L M

PY - 2014/6

Y1 - 2014/6

N2 - Surrogate models are an important part of building energy labelling programs, but these models still present low accuracy, particularly in cooling-dominated climates. The objective of this study was to evaluate the feasibility of using an artificial neural network (ANN) to improve the accuracy of surrogate models for labelling purposes. An ANN was applied to model the building stock of a city in Brazil, based on the results of extensive simulations using the high-resolution building energy simulation program EnergyPlus. Sensitivity and uncertainty analyses were carried out to evaluate the behaviour of the ANN model, and the variations in the best and worst performance for several typologies were analysed in relation to variations in the input parameters and building characteristics. The results obtained indicate that an ANN can represent the interaction between input and output data for a vast and diverse building stock. Sensitivity analysis showed that no single input parameter can be identified as the main factor responsible for the building energy performance. The uncertainty associated with several parameters plays a major role in assessing building energy performance, together with the facade area and the shell-to-floor ratio. The results of this study may have a profound impact as ANNs could be applied in the future to define regulations in many countries, with positive effects on optimizing the energy consumption.

AB - Surrogate models are an important part of building energy labelling programs, but these models still present low accuracy, particularly in cooling-dominated climates. The objective of this study was to evaluate the feasibility of using an artificial neural network (ANN) to improve the accuracy of surrogate models for labelling purposes. An ANN was applied to model the building stock of a city in Brazil, based on the results of extensive simulations using the high-resolution building energy simulation program EnergyPlus. Sensitivity and uncertainty analyses were carried out to evaluate the behaviour of the ANN model, and the variations in the best and worst performance for several typologies were analysed in relation to variations in the input parameters and building characteristics. The results obtained indicate that an ANN can represent the interaction between input and output data for a vast and diverse building stock. Sensitivity analysis showed that no single input parameter can be identified as the main factor responsible for the building energy performance. The uncertainty associated with several parameters plays a major role in assessing building energy performance, together with the facade area and the shell-to-floor ratio. The results of this study may have a profound impact as ANNs could be applied in the future to define regulations in many countries, with positive effects on optimizing the energy consumption.

KW - artificial neural network

KW - building energy simulation

KW - surrogate model

UR - http://www.scopus.com/inward/record.url?scp=84899075409&partnerID=8YFLogxK

U2 - 10.1016/j.enpol.2014.02.001

DO - 10.1016/j.enpol.2014.02.001

M3 - Article

VL - 69

SP - 457

EP - 466

JO - Energy Policy

T2 - Energy Policy

JF - Energy Policy

SN - 0301-4215

ER -