Development of a data-driven predictive model of supply air temperature in an air-handling unit for conserving energy

Goopyo Hong, Byungseon Sean Kim

Research output: Contribution to journalArticle

6 Citations (Scopus)

Abstract

The purpose of this study was to develop a data-driven predictive model that can predict the supply air temperature (SAT) in an air-handling unit (AHU) by using a neural network. A case study was selected, and AHU operational data from December 2015 to November 2016 was collected. A data-driven predictive model was generated through an evolving process that consisted of an initial model, an optimal model, and an adaptive model. In order to develop the optimal model, input variables, the number of neurons and hidden layers, and the period of the training data set were considered. Since AHU data changes over time, an adaptive model, which has the ability to actively cope with constantly changing data, was developed. This adaptive model determined the model with the lowest mean square error (MSE) of the 91 models, which had two hidden layers and sets up a 12-hour test set at every prediction. The adaptive model used recently collected data as training data and utilized the sliding window technique rather than the accumulative data method. Furthermore, additional testing was performed to validate the adaptive model using AHU data from another building. The final adaptive model predicts SAT to a root mean square error (RMSE) of less than 0.6 °C.

Original languageEnglish
Article number407
JournalEnergies
Volume11
Issue number2
DOIs
Publication statusPublished - 2018 Feb

Fingerprint

Predictive Model
Data-driven
Unit
Air
Energy
Temperature
Model
Mean square error
Predict
Sliding Window
Test Set
Neuron
Lowest
Neurons
Roots
Neural Networks
Testing

All Science Journal Classification (ASJC) codes

  • Renewable Energy, Sustainability and the Environment
  • Energy Engineering and Power Technology
  • Energy (miscellaneous)
  • Control and Optimization
  • Electrical and Electronic Engineering

Cite this

@article{c722a9413dfe47049236eed6c8b3d5d5,
title = "Development of a data-driven predictive model of supply air temperature in an air-handling unit for conserving energy",
abstract = "The purpose of this study was to develop a data-driven predictive model that can predict the supply air temperature (SAT) in an air-handling unit (AHU) by using a neural network. A case study was selected, and AHU operational data from December 2015 to November 2016 was collected. A data-driven predictive model was generated through an evolving process that consisted of an initial model, an optimal model, and an adaptive model. In order to develop the optimal model, input variables, the number of neurons and hidden layers, and the period of the training data set were considered. Since AHU data changes over time, an adaptive model, which has the ability to actively cope with constantly changing data, was developed. This adaptive model determined the model with the lowest mean square error (MSE) of the 91 models, which had two hidden layers and sets up a 12-hour test set at every prediction. The adaptive model used recently collected data as training data and utilized the sliding window technique rather than the accumulative data method. Furthermore, additional testing was performed to validate the adaptive model using AHU data from another building. The final adaptive model predicts SAT to a root mean square error (RMSE) of less than 0.6 °C.",
author = "Goopyo Hong and Kim, {Byungseon Sean}",
year = "2018",
month = "2",
doi = "10.3390/en11020407",
language = "English",
volume = "11",
journal = "Energies",
issn = "1996-1073",
publisher = "Multidisciplinary Digital Publishing Institute (MDPI)",
number = "2",

}

Development of a data-driven predictive model of supply air temperature in an air-handling unit for conserving energy. / Hong, Goopyo; Kim, Byungseon Sean.

In: Energies, Vol. 11, No. 2, 407, 02.2018.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Development of a data-driven predictive model of supply air temperature in an air-handling unit for conserving energy

AU - Hong, Goopyo

AU - Kim, Byungseon Sean

PY - 2018/2

Y1 - 2018/2

N2 - The purpose of this study was to develop a data-driven predictive model that can predict the supply air temperature (SAT) in an air-handling unit (AHU) by using a neural network. A case study was selected, and AHU operational data from December 2015 to November 2016 was collected. A data-driven predictive model was generated through an evolving process that consisted of an initial model, an optimal model, and an adaptive model. In order to develop the optimal model, input variables, the number of neurons and hidden layers, and the period of the training data set were considered. Since AHU data changes over time, an adaptive model, which has the ability to actively cope with constantly changing data, was developed. This adaptive model determined the model with the lowest mean square error (MSE) of the 91 models, which had two hidden layers and sets up a 12-hour test set at every prediction. The adaptive model used recently collected data as training data and utilized the sliding window technique rather than the accumulative data method. Furthermore, additional testing was performed to validate the adaptive model using AHU data from another building. The final adaptive model predicts SAT to a root mean square error (RMSE) of less than 0.6 °C.

AB - The purpose of this study was to develop a data-driven predictive model that can predict the supply air temperature (SAT) in an air-handling unit (AHU) by using a neural network. A case study was selected, and AHU operational data from December 2015 to November 2016 was collected. A data-driven predictive model was generated through an evolving process that consisted of an initial model, an optimal model, and an adaptive model. In order to develop the optimal model, input variables, the number of neurons and hidden layers, and the period of the training data set were considered. Since AHU data changes over time, an adaptive model, which has the ability to actively cope with constantly changing data, was developed. This adaptive model determined the model with the lowest mean square error (MSE) of the 91 models, which had two hidden layers and sets up a 12-hour test set at every prediction. The adaptive model used recently collected data as training data and utilized the sliding window technique rather than the accumulative data method. Furthermore, additional testing was performed to validate the adaptive model using AHU data from another building. The final adaptive model predicts SAT to a root mean square error (RMSE) of less than 0.6 °C.

UR - http://www.scopus.com/inward/record.url?scp=85045383607&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85045383607&partnerID=8YFLogxK

U2 - 10.3390/en11020407

DO - 10.3390/en11020407

M3 - Article

AN - SCOPUS:85045383607

VL - 11

JO - Energies

JF - Energies

SN - 1996-1073

IS - 2

M1 - 407

ER -