Prediction models and control algorithms for predictive applications of setback temperature in cooling systems

Jin Woo Moon, Younju Yoon, Young Hoon Jeon, Sooyoung Kim

Research output: Contribution to journalArticle

6 Citations (Scopus)

Abstract

In this study, a temperature control algorithm was developed to apply a setback temperature predictively for the cooling system of a residential building during occupied periods by residents. An artificial neural network (ANN) model was developed to determine the required time for increasing the current indoor temperature to the setback temperature. This study involved three phases: development of the initial ANN-based prediction model, optimization and testing of the initial model, and development and testing of three control algorithms. The development and performance testing of the model and algorithm were conducted using TRNSYS and MATLAB. Through the development and optimization process, the final ANN model employed indoor temperature and the temperature difference between the current and target setback temperature as two input neurons. The optimal number of hidden layers, number of neurons, learning rate, and moment were determined to be 4, 9, 0.6, and 0.9, respectively. The tangent–sigmoid and pure-linear transfer function was used in the hidden and output neurons, respectively. The ANN model used 100 training data sets with sliding-window method for data management. Levenberg-Marquart training method was employed for model training. The optimized model had a prediction accuracy of 0.9097 root mean square errors when compared with the simulated results. Employing the ANN model, ANN-based algorithms maintained indoor temperatures better within target ranges. Compared to the conventional algorithm, the ANN-based algorithms reduced the duration of time, in which the indoor temperature was out of the targeted temperature range, as much as 56 and 75 min, respectively. In addition, two ANN-based algorithms removed less heat from indoor space as much as 1.06% and 1.26%. Thus, the applicability of the ANN model and the algorithm presented their potential to be applied for more effective thermal conditioning with reduced energy consumption.

Original languageEnglish
Pages (from-to)1290-1302
Number of pages13
JournalApplied Thermal Engineering
Volume113
DOIs
Publication statusPublished - 2017 Feb 25

Fingerprint

Cooling systems
Neural networks
Temperature
Neurons
Testing
Temperature control
Mean square error
Information management
MATLAB
Transfer functions
Energy utilization

All Science Journal Classification (ASJC) codes

  • Energy Engineering and Power Technology
  • Industrial and Manufacturing Engineering

Cite this

@article{30bedf4b60854c62af70d083975c147f,
title = "Prediction models and control algorithms for predictive applications of setback temperature in cooling systems",
abstract = "In this study, a temperature control algorithm was developed to apply a setback temperature predictively for the cooling system of a residential building during occupied periods by residents. An artificial neural network (ANN) model was developed to determine the required time for increasing the current indoor temperature to the setback temperature. This study involved three phases: development of the initial ANN-based prediction model, optimization and testing of the initial model, and development and testing of three control algorithms. The development and performance testing of the model and algorithm were conducted using TRNSYS and MATLAB. Through the development and optimization process, the final ANN model employed indoor temperature and the temperature difference between the current and target setback temperature as two input neurons. The optimal number of hidden layers, number of neurons, learning rate, and moment were determined to be 4, 9, 0.6, and 0.9, respectively. The tangent–sigmoid and pure-linear transfer function was used in the hidden and output neurons, respectively. The ANN model used 100 training data sets with sliding-window method for data management. Levenberg-Marquart training method was employed for model training. The optimized model had a prediction accuracy of 0.9097 root mean square errors when compared with the simulated results. Employing the ANN model, ANN-based algorithms maintained indoor temperatures better within target ranges. Compared to the conventional algorithm, the ANN-based algorithms reduced the duration of time, in which the indoor temperature was out of the targeted temperature range, as much as 56 and 75 min, respectively. In addition, two ANN-based algorithms removed less heat from indoor space as much as 1.06{\%} and 1.26{\%}. Thus, the applicability of the ANN model and the algorithm presented their potential to be applied for more effective thermal conditioning with reduced energy consumption.",
author = "Moon, {Jin Woo} and Younju Yoon and Jeon, {Young Hoon} and Sooyoung Kim",
year = "2017",
month = "2",
day = "25",
doi = "10.1016/j.applthermaleng.2016.11.087",
language = "English",
volume = "113",
pages = "1290--1302",
journal = "Applied Thermal Engineering",
issn = "1359-4311",
publisher = "Elsevier Limited",

}

Prediction models and control algorithms for predictive applications of setback temperature in cooling systems. / Moon, Jin Woo; Yoon, Younju; Jeon, Young Hoon; Kim, Sooyoung.

In: Applied Thermal Engineering, Vol. 113, 25.02.2017, p. 1290-1302.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Prediction models and control algorithms for predictive applications of setback temperature in cooling systems

AU - Moon, Jin Woo

AU - Yoon, Younju

AU - Jeon, Young Hoon

AU - Kim, Sooyoung

PY - 2017/2/25

Y1 - 2017/2/25

N2 - In this study, a temperature control algorithm was developed to apply a setback temperature predictively for the cooling system of a residential building during occupied periods by residents. An artificial neural network (ANN) model was developed to determine the required time for increasing the current indoor temperature to the setback temperature. This study involved three phases: development of the initial ANN-based prediction model, optimization and testing of the initial model, and development and testing of three control algorithms. The development and performance testing of the model and algorithm were conducted using TRNSYS and MATLAB. Through the development and optimization process, the final ANN model employed indoor temperature and the temperature difference between the current and target setback temperature as two input neurons. The optimal number of hidden layers, number of neurons, learning rate, and moment were determined to be 4, 9, 0.6, and 0.9, respectively. The tangent–sigmoid and pure-linear transfer function was used in the hidden and output neurons, respectively. The ANN model used 100 training data sets with sliding-window method for data management. Levenberg-Marquart training method was employed for model training. The optimized model had a prediction accuracy of 0.9097 root mean square errors when compared with the simulated results. Employing the ANN model, ANN-based algorithms maintained indoor temperatures better within target ranges. Compared to the conventional algorithm, the ANN-based algorithms reduced the duration of time, in which the indoor temperature was out of the targeted temperature range, as much as 56 and 75 min, respectively. In addition, two ANN-based algorithms removed less heat from indoor space as much as 1.06% and 1.26%. Thus, the applicability of the ANN model and the algorithm presented their potential to be applied for more effective thermal conditioning with reduced energy consumption.

AB - In this study, a temperature control algorithm was developed to apply a setback temperature predictively for the cooling system of a residential building during occupied periods by residents. An artificial neural network (ANN) model was developed to determine the required time for increasing the current indoor temperature to the setback temperature. This study involved three phases: development of the initial ANN-based prediction model, optimization and testing of the initial model, and development and testing of three control algorithms. The development and performance testing of the model and algorithm were conducted using TRNSYS and MATLAB. Through the development and optimization process, the final ANN model employed indoor temperature and the temperature difference between the current and target setback temperature as two input neurons. The optimal number of hidden layers, number of neurons, learning rate, and moment were determined to be 4, 9, 0.6, and 0.9, respectively. The tangent–sigmoid and pure-linear transfer function was used in the hidden and output neurons, respectively. The ANN model used 100 training data sets with sliding-window method for data management. Levenberg-Marquart training method was employed for model training. The optimized model had a prediction accuracy of 0.9097 root mean square errors when compared with the simulated results. Employing the ANN model, ANN-based algorithms maintained indoor temperatures better within target ranges. Compared to the conventional algorithm, the ANN-based algorithms reduced the duration of time, in which the indoor temperature was out of the targeted temperature range, as much as 56 and 75 min, respectively. In addition, two ANN-based algorithms removed less heat from indoor space as much as 1.06% and 1.26%. Thus, the applicability of the ANN model and the algorithm presented their potential to be applied for more effective thermal conditioning with reduced energy consumption.

UR - http://www.scopus.com/inward/record.url?scp=84997769522&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84997769522&partnerID=8YFLogxK

U2 - 10.1016/j.applthermaleng.2016.11.087

DO - 10.1016/j.applthermaleng.2016.11.087

M3 - Article

AN - SCOPUS:84997769522

VL - 113

SP - 1290

EP - 1302

JO - Applied Thermal Engineering

JF - Applied Thermal Engineering

SN - 1359-4311

ER -