Coupling stochastic methods and detailed dynamic simulation programs for model calibration: two preliminary case studies

Dynamic simulation programs for energy modelling have reached a high level of detail and accuracy in representing the main phenomena that determine energy performances of buildings. Many of these programs have been subjected to numerous validation studies which have demonstrated their capability of representing reality adequately if the correct inputs are available. However, that is rarely the case and often many input variables are unknown or subject to high uncertainty making predictions quite different from reality.
To overcome this issue, probabilistic models can be used, in order to learn from field data and use such updated knowledge to improve physical models. This
work proposes a framework to apply such concept and shows some interesting and promising results for two simple preliminary case studies.