Monday, 27 June 2016: 8:00 AM-9:30 AM
Fundamentals and Applications
Chair:
Chris Balbach, P.E., Performance Systems Development
Technical Committee: 04.07 Energy Calculations
Although building energy modeling has been common for many years, tools that support large-scale modeling analysis by leveraging vast cloud computing power are now both affordable and accessible. While these approaches make it easy to analyze tens of thousands of model variants, they may not take the shortest path to lead users to the answers they seek. In this session, presenters share case studies involving large scale modeling and results analysis. Attendees learn how to effectively efficiently design a large scale simulation study.
1 Design of Experiments: Statistical Confidence with Fewer Simulations
 |
New Joshua, Ph.D., Oak Ridge National Laboratory |
The increasing use of parametric ensembles with building energy models to study sensitivity and accommodate uncertainty has the potential to greatly inform energy studies, but can be mitigated by low statistical confidence from poor experimental design. In this talk, we present the tradeoffs between common statistical approaches to design of experiments and how they can be used in cloud or supercomputing resources.
2 Exercising Occam's Razor: Sensitivity Screening Methods as Applied to Building Energy Models
 |
David Bosworth, BUILDlab, LLC |
Building Energy Models are complex and have a lot of inputs, and a minority of the inputs them have a significant effect on a result. Sensitivity screening tools are designed to be computationally cheap (requiring a relatively small number of simulations) as they rank the model inputs in order of their influence on a particular output. The goal is to identify which model inputs require the modelers attention, and which can be ignored. Speakers in this session discuss and demonstrate the use of the Morris Method, a commonly used sensitivity screening algorithm.
3 How to Do Energy Model Uncertainty Analysis with Correlated Input Variables
 |
Ralph Muehleisen, Ph.D., P.E., Argonne National Laboratory
|
|
|
Joshua Bergerson, Ph.D., Argonne National Laboratory |
Energy modelers are starting to try to quantify the uncertainty in their energy models. The methods for estimating uncertainty when inputs are independent are fairly well known. However, in the case of buildings, many inputs are not independent. In particular, occupant related loads such as plug loads, lighting loads, and occupant heat loads are known to be well correlated. Uncertainties of correlated variables can be propagated if joint probability distributions are used and the joint distributions are properly sampled. In this paper, the selection and sampling of joint distributions for uncertainty with correlated inputs is discussed.