Elsevier

Energy

Volume 20, Issue 12, December 1995, Pages 1291-1301
Energy

Validation of an algorithm to disaggregate whole-building hourly electrical load into end uses

https://doi.org/10.1016/0360-5442(95)00033-DGet rights and content

Abstract

We have developed an algorithm to disaggregate short-interval (hourly) whole-building electrical load into major end uses. Hourly load data, hourly load-temperature regression coefficients and simulation end-use results comprise the algorithm input. The algorithm produces hourly load profiles for air conditioning, lighting, fans and pumps, and miscellaneous loads. Measured data from two end-use metered buildings (an office and a retail store) have been used to validate the algorithm. For the retail store, the algorithm estimates of hourly end use compare remarkably well with the monitored end-use data (average error of less than 5% during daytime operation). For the office building, the algorithm gives a consistent bias of about 12 and 27% in overestimating the HVAC and lighting electric loads, respectively, at the expense of underestimating the miscellaenous load by 35%. Results may be attributed to the presence of inconsistencies between office audit information and measured end-use data. A three-fold difference between the auditor's estimate for miscellaneous energy use and the metered amount has been found. The validation, however, indicates great promise for application of the algorithm to whole-building load data for obtaining reliable end-use data.

References (19)

  • D. Flora et al.

    Methods to Obtain Building Energy Performance Data: Are Energy Management Systems Promising Sources?

    D. Flora et al.

    Lawrence Berkeley Laboratory Report LBL-20897

    (1986)
  • P. Le Coniac et al.

    Energy Management Systems as a Source of Building Energy Performance Data

  • J.A. Heidell et al.

    Commercial Building End-Use Energy Metering Inventory

    Pacific Northwest Laboratory Report PNL-5027

    (1985)
  • B.L. Gardiner

    The CAL-BECA Project Part 1: New California Nonresidential Buildings

    Lawrence Bekeley Laboratory Report LBL-19244

    (1985)
  • J.D. Burch

    Approaches to Analyzing the Thermal Performance of Commercial Buildings

  • K. Train et al.

    Energy—The International Journal

    (1985)
  • W.T. Snyder et al.

    The Energy Audit—What it is, How to Conduct it, How to Use the Results

  • M.A. Meal et al.

    ASHRAE Trans.

    (1985)
There are more references available in the full text version of this article.

Cited by (23)

  • A simultaneous calibration and parameter ranking method for building energy models

    2017, Applied Energy
    Citation Excerpt :

    In general, the calibration methods can be categorized as manual and automatic calibrations [12]. Manual calibration can be implemented through various ways, such as through the characterisation of the physical and operational properties of existing buildings [17], graphical representation of building data or statistical indices [18], parameter reductions [19,20], and data disaggregation [21]. The manual calibration methods are usually dependent on expert knowledge and judgement, which can be prone to error.

  • Evaluation of “Autotune” calibration against manual calibration of building energy models

    2016, Applied Energy
    Citation Excerpt :

    The relative development time of these models is short, although they can be brittle and require re-training to accommodate small changes in a building. Examples include models created by traditional regression [18–20], artificial neural networks [21–23], or support vector machines [24–26]. Grey-box models often use parameters identified from physical systems, can use decision tree [27,28] or Fourier series techniques [29–31], and can account for changes caused by simplified input parameters.

  • A bottom-up and procedural calibration method for building energy simulation models based on hourly electricity submetering data

    2015, Energy
    Citation Excerpt :

    1) Black-box models (data-driven models) are simple mathematical or statistical models resulting from long-term historical data training. These models have high requirements for the quality and quantity of data, but lack physical meaning, and they include traditional regression models [6–8], ANN (artificial neural network) models [9–11] and SVM (support vector machine) models [12–14]. 2) Grey-box models differ from black-box approaches in that they use certain parameters identified from physical system models, such as decision tree models [15,16] and Fourier series models [17–19].

View all citing articles on Scopus
View full text