Bayesian verification of an energy conservation measure

We have tested this method after installing, in March 2016, a model-predictive controller for space heating for a medium-sized (about 2700 m2) office building in Switzerland. The baseline was established from historic oil tank refill records and climate data from a nearby weather station. The total heat loss coefficient of the building was assessed 46 days after the installation and found to have decreased by 31.9%, with 17.8 percentage points standard error.