View in EDS HTML Full Text PDF Full Text

Investigation of Different Feature Selection Methods for Virtual Sensors of District Heating Systems.

Saved in:
Bibliographic Details
Title: Investigation of Different Feature Selection Methods for Virtual Sensors of District Heating Systems.
Authors: Sha, Haohan1 (AUTHOR), Xu, Zheng2 (AUTHOR), Gao, Junjie3 (AUTHOR), Yu, Hongrui1 (AUTHOR), Shi, Zhigang1,2 (AUTHOR), Tu, Jin2,3 (AUTHOR), Qiu, Hang2 (AUTHOR)
Source: Energies (19961073). May2026, Vol. 19 Issue 9, p2062. 28p.
Subject Terms: *Feature selection, *Machine learning, *Regression trees, *Heating from central stations, *Temperature control
Abstract: District heating systems play a critical role in urban energy supply; however, secondary networks suffer from frequent sensor failures that undermine thermal balance control. The development of virtual sensors to estimate return-water temperatures offers a promising solution to this challenge. This study investigates the performance of different feature selection methods for developing virtual sensors. The investigated feature selection methods include two engineering experience-based methods, one embedded method, one wrapped-based method, and two filter methods. Using operational data from a real secondary district heating network over an entire heating season, the embedded method's performance is investigated, and an appropriate machine learning algorithm, paired with the wrapped and filter methods, is selected. For the filter methods, the paper additionally examines the differences between the rank-based and threshold-based filter implementations. The performance of the wrapped and filter methods on accuracy, computational cost, and sensitivity to data volume was compared. The results indicate that the embedded method exhibits relatively unstable performance under the engineering experience-based baselines, but the gradient tree boosting (GTB) method demonstrates better performance in both accuracy and stability. Further tests combining GTB with wrapped and filter methods revealed that both the filter and wrapped methods show an acceptable performance in terms of accuracy. The mean RMSE of both filter and wrapped methods consistently ranges from 0.75 °C to 0.8 °C when the selected feature is more than 6. However, the wrapped method exhibits a higher computational cost and is more sensitive to data volume. The training time of the wrapped method is approximately 136 times that of the fastest filter method. Considering overall performance, the combination of GTB with an HSIC indicator, employing either the rank-based selection of the top 9 features or the threshold-based feature selection, is recommended. These findings provide methodological guidance for the development of virtual sensors in district heating systems. [ABSTRACT FROM AUTHOR]
Database: Energy & Power Source
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
Text:
  Availability: 1
Header DbId: enr
DbLabel: Energy & Power Source
An: 193715958
AccessLevel: 6
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Investigation of Different Feature Selection Methods for Virtual Sensors of District Heating Systems.
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Sha%2C+Haohan%22">Sha, Haohan</searchLink><relatesTo>1</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Xu%2C+Zheng%22">Xu, Zheng</searchLink><relatesTo>2</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Gao%2C+Junjie%22">Gao, Junjie</searchLink><relatesTo>3</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Yu%2C+Hongrui%22">Yu, Hongrui</searchLink><relatesTo>1</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Shi%2C+Zhigang%22">Shi, Zhigang</searchLink><relatesTo>1,2</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Tu%2C+Jin%22">Tu, Jin</searchLink><relatesTo>2,3</relatesTo> (AUTHOR)<br /><searchLink fieldCode="AR" term="%22Qiu%2C+Hang%22">Qiu, Hang</searchLink><relatesTo>2</relatesTo> (AUTHOR)
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="JN" term="%22Energies+%2819961073%29%22">Energies (19961073)</searchLink>. May2026, Vol. 19 Issue 9, p2062. 28p.
– Name: Subject
  Label: Subject Terms
  Group: Su
  Data: *<searchLink fieldCode="DE" term="%22Feature+selection%22">Feature selection</searchLink><br />*<searchLink fieldCode="DE" term="%22Machine+learning%22">Machine learning</searchLink><br />*<searchLink fieldCode="DE" term="%22Regression+trees%22">Regression trees</searchLink><br />*<searchLink fieldCode="DE" term="%22Heating+from+central+stations%22">Heating from central stations</searchLink><br />*<searchLink fieldCode="DE" term="%22Temperature+control%22">Temperature control</searchLink>
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: District heating systems play a critical role in urban energy supply; however, secondary networks suffer from frequent sensor failures that undermine thermal balance control. The development of virtual sensors to estimate return-water temperatures offers a promising solution to this challenge. This study investigates the performance of different feature selection methods for developing virtual sensors. The investigated feature selection methods include two engineering experience-based methods, one embedded method, one wrapped-based method, and two filter methods. Using operational data from a real secondary district heating network over an entire heating season, the embedded method's performance is investigated, and an appropriate machine learning algorithm, paired with the wrapped and filter methods, is selected. For the filter methods, the paper additionally examines the differences between the rank-based and threshold-based filter implementations. The performance of the wrapped and filter methods on accuracy, computational cost, and sensitivity to data volume was compared. The results indicate that the embedded method exhibits relatively unstable performance under the engineering experience-based baselines, but the gradient tree boosting (GTB) method demonstrates better performance in both accuracy and stability. Further tests combining GTB with wrapped and filter methods revealed that both the filter and wrapped methods show an acceptable performance in terms of accuracy. The mean RMSE of both filter and wrapped methods consistently ranges from 0.75 °C to 0.8 °C when the selected feature is more than 6. However, the wrapped method exhibits a higher computational cost and is more sensitive to data volume. The training time of the wrapped method is approximately 136 times that of the fastest filter method. Considering overall performance, the combination of GTB with an HSIC indicator, employing either the rank-based selection of the top 9 features or the threshold-based feature selection, is recommended. These findings provide methodological guidance for the development of virtual sensors in district heating systems. [ABSTRACT FROM AUTHOR]
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=enr&AN=193715958
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.3390/en19092062
    Languages:
      – Code: eng
        Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 28
        StartPage: 2062
    Subjects:
      – SubjectFull: Feature selection
        Type: general
      – SubjectFull: Machine learning
        Type: general
      – SubjectFull: Regression trees
        Type: general
      – SubjectFull: Heating from central stations
        Type: general
      – SubjectFull: Temperature control
        Type: general
    Titles:
      – TitleFull: Investigation of Different Feature Selection Methods for Virtual Sensors of District Heating Systems.
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Sha, Haohan
      – PersonEntity:
          Name:
            NameFull: Xu, Zheng
      – PersonEntity:
          Name:
            NameFull: Gao, Junjie
      – PersonEntity:
          Name:
            NameFull: Yu, Hongrui
      – PersonEntity:
          Name:
            NameFull: Shi, Zhigang
      – PersonEntity:
          Name:
            NameFull: Tu, Jin
      – PersonEntity:
          Name:
            NameFull: Qiu, Hang
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 05
              Text: May2026
              Type: published
              Y: 2026
          Identifiers:
            – Type: issn-print
              Value: 19961073
          Numbering:
            – Type: volume
              Value: 19
            – Type: issue
              Value: 9
          Titles:
            – TitleFull: Energies (19961073)
              Type: main
ResultId 1