A Protocol-Based, Inverse-Model-Driven Methodology for Building Auditing and Forward Energy Model Formulation

Open Access
Lin, Bo
Graduate Program:
Architectural Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
August 28, 2015
Committee Members:
  • James Freihaut, Dissertation Advisor
  • Chinemelu Jidenka Anumba, Committee Member
  • Stephen James Treado, Committee Member
  • Runze Li, Committee Member
  • Inverse model
  • energy model
  • calibration
  • measurement & verification
  • sub-metering
  • variable selection
  • benchmark
  • audit
This dissertation demonstrates two methods to formulate inverse regression models of a building’s as operated energy utilization. Two advanced and robust variable selection techniques -Least Absolute Shrinkage and Selection Operator (LASSO), and Variable Selection though Smoothly Clipped Absolute Deviation Penalty (SCAD) - are implemented as part of the regression based model development. It is established that independent variables determined to be important in influencing building energy models use should be metered prior to the building retrofit or audit. A medium sized office building developed by U.S Department of Energy (DOE), also known as the DOE commercial reference buildings is utilized to develop the model formulation methodology. The building represents an average condition of numerous buildings in the US medium size office building sector. The “typical” office building condition is established as a baseline model case for the analysis. Input data variation, models generations and independent variable selections are determined using the baseline building through the use of EnergyPlus and Matlab tools. Multi-variant regression models are formulated for whole building energy utilization, the cooling system and heating sub-system utilization. Statistical indices indicate that the models developed though the two methods are valid and effective in confidently predicting building energy utilization. Though the LASSO and SCAD variable selection techniques, key variables that are statistically significant to building energy use are identified. Those variables should be given priority in any building monitoring plan as their characterization with assist in energy utilization analyses to continuously improve building energy performance. In addition to establishing inverse model formulation though variable selection techniques, this investigation developed a variation bin method to identify “inefficient” sub-system utilization in building energy use. The idea of bin method is to compare energy variation of sub-systems’ data by excluding impacts from ambient weather conditions. The advantages of this method are obvious. Firstly, it is easy and straightforward to apply. Secondly, it can be applied to the entire metered sub-systems data. The methodology is used to formulate as-operated, low variance forward building energy models. Forward building energy models enables parametric analysis to evaluate design or renovation strategies. It allows energy efficiency estimates that can be achieved by implementation of energy standards or codes and assists in application for design certifications. As-operated building energy models are often developed to assist energy conservation measures (ECMs) design and evaluation. The methodology developed in this investigation results in an actual data based, sub-system focused energy model. The methodology avoids trial and error, time consuming, labor intensive approaches. The model is considered calibrated when simulated heating, cooling, plug and lighting sub-systems end use and whole building energy use meet predefined statistical criteria. Two medium size office buildings in Pennsylvania are selected as case studies and show that the method is valid and efficient. Monthly and weekday daily comparisons indicate that simulated results satisfy statistical criteria and models are considered as calibrated.