Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate Modeling

Surrogate modeling technology is used to create lightweight analogs of resource- and calculation-intensive software, provided that the problem can be reduced to the regression problem. In this article, we construct a surrogate model for predicting annual energy consumption using the open-source Ener...

Full description

Saved in:
Bibliographic Details
Main Authors: Leonid Legashev, Sergey Tolmachev, Irina Bolodurina, Alexander Shukhman, Lyubov Grishina
Format: Article
Language:English
Published: MDPI AG 2024-12-01
Series:Modelling
Subjects:
Online Access:https://www.mdpi.com/2673-3951/5/4/106
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1846103540571832320
author Leonid Legashev
Sergey Tolmachev
Irina Bolodurina
Alexander Shukhman
Lyubov Grishina
author_facet Leonid Legashev
Sergey Tolmachev
Irina Bolodurina
Alexander Shukhman
Lyubov Grishina
author_sort Leonid Legashev
collection DOAJ
description Surrogate modeling technology is used to create lightweight analogs of resource- and calculation-intensive software, provided that the problem can be reduced to the regression problem. In this article, we construct a surrogate model for predicting annual energy consumption using the open-source EnergyPlus software and various sampling techniques. A general algorithm for an error-based adaptive sampling technique to build the surrogate model is presented. The best results were shown by the composite Mixed Sampling method with a data refining window the size of 70% and a LightGBM regression model. The best attained metrics values are as follows: MSE = 7.76, RMSE = 1.47, MAE = 0.98 and R<sup>2</sup> = 0.99. For a small number of iterations, an error-based adaptive sampling technique with hyperparameter tuning is preferable to the static sampling approach. For a large number of iterations, both techniques show approximately good predictive results of the built surrogate model. After hyperparameter tuning was performed, the average value of the MSE metric decreased from 43.43 to 7.76. A gas thickness feature greater than 0.015 had no positive effect on energy-saving optimization. For temperatures on a summer day of 30 degrees and above, there was a sharp increase in energy consumption. The maximum dry bulb temperature on a winter and summer day and the wind speed on a winter day were the most important features of the built surrogate model with 492, 483 and 443 gain values of the feature importance method, respectively.
format Article
id doaj-art-e1f1649205804dfea19bc98d96c98c14
institution Kabale University
issn 2673-3951
language English
publishDate 2024-12-01
publisher MDPI AG
record_format Article
series Modelling
spelling doaj-art-e1f1649205804dfea19bc98d96c98c142024-12-27T14:42:09ZengMDPI AGModelling2673-39512024-12-01542051207410.3390/modelling5040106Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate ModelingLeonid Legashev0Sergey Tolmachev1Irina Bolodurina2Alexander Shukhman3Lyubov Grishina4Research Center “Strong Artificial Intelligence in Industry”, ITMO University, Kronverksky Pr. 49, bldg. A, St. Petersburg 197101, RussiaResearch Center “Strong Artificial Intelligence in Industry”, ITMO University, Kronverksky Pr. 49, bldg. A, St. Petersburg 197101, RussiaResearch Center “Strong Artificial Intelligence in Industry”, ITMO University, Kronverksky Pr. 49, bldg. A, St. Petersburg 197101, RussiaResearch Center “Strong Artificial Intelligence in Industry”, ITMO University, Kronverksky Pr. 49, bldg. A, St. Petersburg 197101, RussiaResearch Center “Strong Artificial Intelligence in Industry”, ITMO University, Kronverksky Pr. 49, bldg. A, St. Petersburg 197101, RussiaSurrogate modeling technology is used to create lightweight analogs of resource- and calculation-intensive software, provided that the problem can be reduced to the regression problem. In this article, we construct a surrogate model for predicting annual energy consumption using the open-source EnergyPlus software and various sampling techniques. A general algorithm for an error-based adaptive sampling technique to build the surrogate model is presented. The best results were shown by the composite Mixed Sampling method with a data refining window the size of 70% and a LightGBM regression model. The best attained metrics values are as follows: MSE = 7.76, RMSE = 1.47, MAE = 0.98 and R<sup>2</sup> = 0.99. For a small number of iterations, an error-based adaptive sampling technique with hyperparameter tuning is preferable to the static sampling approach. For a large number of iterations, both techniques show approximately good predictive results of the built surrogate model. After hyperparameter tuning was performed, the average value of the MSE metric decreased from 43.43 to 7.76. A gas thickness feature greater than 0.015 had no positive effect on energy-saving optimization. For temperatures on a summer day of 30 degrees and above, there was a sharp increase in energy consumption. The maximum dry bulb temperature on a winter and summer day and the wind speed on a winter day were the most important features of the built surrogate model with 492, 483 and 443 gain values of the feature importance method, respectively.https://www.mdpi.com/2673-3951/5/4/106surrogate modelingenergy optimizationEnergyPlusdesign of experimentprediction
spellingShingle Leonid Legashev
Sergey Tolmachev
Irina Bolodurina
Alexander Shukhman
Lyubov Grishina
Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate Modeling
Modelling
surrogate modeling
energy optimization
EnergyPlus
design of experiment
prediction
title Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate Modeling
title_full Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate Modeling
title_fullStr Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate Modeling
title_full_unstemmed Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate Modeling
title_short Investigation into the Hyperparameters of Error-Based Adaptive Sampling Approach for Surrogate Modeling
title_sort investigation into the hyperparameters of error based adaptive sampling approach for surrogate modeling
topic surrogate modeling
energy optimization
EnergyPlus
design of experiment
prediction
url https://www.mdpi.com/2673-3951/5/4/106
work_keys_str_mv AT leonidlegashev investigationintothehyperparametersoferrorbasedadaptivesamplingapproachforsurrogatemodeling
AT sergeytolmachev investigationintothehyperparametersoferrorbasedadaptivesamplingapproachforsurrogatemodeling
AT irinabolodurina investigationintothehyperparametersoferrorbasedadaptivesamplingapproachforsurrogatemodeling
AT alexandershukhman investigationintothehyperparametersoferrorbasedadaptivesamplingapproachforsurrogatemodeling
AT lyubovgrishina investigationintothehyperparametersoferrorbasedadaptivesamplingapproachforsurrogatemodeling