Skip to main content

Bayesian Measurement and Verification

Welcome

This website promotes the use of Bayesian data analysis for building energy use monitoring, especially in the context of Measurement and Verification (M&V) methods.

The target audience of the book are energy experts who may only have a moderate background on statistics and applied mathematics. One of our main arguments for promoting Bayesian data analysis is that it is both universally applicable, and accessible. Our aim is to give all building energy practicioners the tools to reliably assess savings following energy conservation measures, along with their uncertainty.

This guide assumes that the reader already knows the core statistical concepts for M&V: errors and uncertainty, regression basics, and model validation metrics. We will however emphasise some of these terms in order to illustrate the specifics of Bayesian methods.

The first part of the website introduces the reader to the main concepts of measurement and verification, and Bayesian data analysis. The amount of theoretical background is voluntarily kept short in order to keep the focus on practical examples. References are included for readers who wish to learn more.

The second part are tutorials on whole-building energy monitoring and M&V, referred to as “Option C” in the IPMVP documentation. This series of three tutorials gradually illustrate the advantages of Bayesian data analysis for M&V:

  • A first tutorial serves as an introduction to the Stan platform, and shows that Bayesian analysis obtains the same uncertainty assessments as classical methods, when analytical solutions are available.
  • Once the reader is reassured on the accuracy of our method, a second tutorial shows how effortlessly it extends to more complex model structures, where analytical solutions may no longer evaluate uncertainty analytically.
  • A third tutorial addresses the issue of residuals autocorrelation, which often arises after model calibration. We show how to overcome this issue with Bayesian models.

The third part of the website is still under construction and will address IPMVP Option D and the so-called Bayesian calibration of building energy models.

The content and how to use it

Most content is displayed from R notebooks which allow displaying the full code for the calculations, their results and explanations. In order to apply the methods to your own use cases, you should first:

We use R and Stan for all analyses, but all content is also adaptable to other environments the reader may be more familiar with: Python, Julia or Matlab. We mention these alternatives here.

Additionally, if you wish to replicate the results shown in these tutorials, you may either download the full book from its repository and run the R markdown files matching the chapters; or only download the data sets and copy/paste the code in your own R files.

About

I am Simon Rouchier, lecturer at the Université Savoie Mont Blanc, Chambéry, France.

Licence Creative Commons
Ce(tte) œuvre est mise à disposition selon les termes de la Licence Creative Commons Attribution - Pas d’Utilisation Commerciale - Partage dans les Mêmes Conditions 4.0 International.