DE-13-008 -- Evolutionary Tuning of Building Models to Monthly Electrical Consumption PDF

Building energy models of existing buildings are unreliableunless calibrated so they correlate well with actual energyuse. Calibrating models is costly because it is currently an artthat requires significant manual effort by an experienced andskilled professional. An automated methodology could significantlydecrease this cost and facilitate greater adoption ofenergy simulation capabilities into the marketplace. TheAutotuneproject is a novel methodology that leverages supercomputing,large databases of simulations, and machine learningto allow automatic calibration of simulations that matchmeasured experimental data. This paper shares initial resultsfrom the automated methodology on commodity hardwareapplied to the calibration of building energy models (BEM) forEnergyPlus (E+) to provide error rates, as measured by thesumof absolute error, for matching monthly load and electricaldata froma highly instrumented and automated ZEBRAllianceresearch home.