Bayesian evaluation of energy conservation measures: a case study with a model-predictive controller for space heating on a commercial building

Abstract We have developed a method, based on Bayesian statistics, to evaluate the efficacy of energy conservation measures on existing buildings. Unlike conventional methods, our method uses existing historic utility bills and climate data to establish a baseline. Comparing this baseline with the measured post-retrofit energy consumption yields the estimated energy savings, including their uncertainties. We have applied this method after installing, in March 2016, a commercial model-predictive controller for space heating on a medium-sized office building in Switzerland. The baseline was established from historic oil tank refill records. 58 days after installing the controller, the energy efficiency of the building had improved by 33.1%, with 19.0 percentage points standard error.