Automatically calibrating a probabilistic graphical model of building energy consumption

We introduce a framework and proof of concept for estimating building energy consumption that probabilistically combines a model of building physics with observed occupancy and detailed operations data, automatically learning a physically plausible model of the energy consumption. Our framework has several desirable properties: data about one building can automatically be used to improve energy use estimates for other similar buildings; input fields can be left blank or specified approximately; and the output of our model is not only an estimate of energy usage, but a probability distribution over possible values. We describe an initial implementation of our framework and present experimental results showing that this is a promising direction for future building simulation research.