Phil Price, Sustainable Energy Systems Group, Lawrence Berkeley National Lab
"How much energy did I save by changing the operation of my building yesterday?" That turns out to be a very hard question to answer: you need to know how much energy you would have used under normal operations (the "baseline"), a number you can predict but not measure. In this talk we focus specifically on electrical energy ("electric load") in commercial buildings. Often the load can be broken down into several components that are superimposed on each other: a recurring weekly pattern, an effect of outdoor air temperature, and so on. Some buildings have patterns that are stable over many weeks, in which case even very simple statistical approaches can quantify these patterns and a predicted baseline can be accurate. But most buildings change somewhat from week to week and month to month: the hours of operation change, or more lights are left on at night, or more people start working late on Thursdays. In this talk I discuss statistical models that try to quantify the changes overall to correctly determine which components have changed and by how much. The resulting models have enormous numbers of parameters, but recent improvements in statistical modeling software make them tractable (just barely). I will show results for some real building data and explain why I think this type of model represents the future for this application and many others.
Presentation attached below.
Past talks >