Calibrating Perez model coefficients using subset simulation

The Perez sky diffuse model has been validated in many locations, however, studies have shown that the precision of estimation can be improved through localization of the model coefficients. This paper studies the effect of tuning Perez irradiance coefficients based on local information for estimating the incident solar radiation on tilted surfaces. The added value of a calibrated Perez model is highlighted by evaluating the heating and cooling energy needs of an office building. Calibration is performed by using Subset Simulation, i.e. a sampling technique based on Markov Chain Monte Carlo. The versatility of the Subset Simulation technique allows for handling multivariant calibration problems and is therefore suitable for finetuning Perez irradiance coefficients. Measurements of global horizontal, direct normal and diffuse horizontal solar irradiation, as well as the global solar irradiation on 90°, 30° and 15° tilted surfaces form the basis of the calibration. It is found that in the studied location, the default Perez model overestimates the incident solar radiation on vertical surfaces facing south. The calibrated coefficients are then embedded in the source code of EnergyPlus. Simulations with a reference office show that the effect of calibrating Perez coefficients can be significant, because it leads to approximately 12% difference in predicted annual cooling energy use and 9% difference in peak cooling loads, respectively.