Uncertainty Quantification

Models of most real-world systems consist of large numbers of uncertain parameters, and often lack accurate description of some key physics. Uncertainty quantification is useful in understanding how the uncertainty in model description and inputs propagates to the outputs or quantities of interest. It is also useful in calibrating models to observation data and to optimize performance in the face of uncertainty.

We have recently used quasi-random sampling based techniques to develop tools for automatic calibration of a black-box model. The application that we are considering is building energy models using tools such as EnergyPlus and TRNSYS. These models typically have hundreds of uncertain parameters that are painstakingly calibrated by domain experts using their experience and intuition.

This is an active area of research and updates will be posted in the future.

S. Peles, S. Ahuja and S. Narayanan. Uncertainty Quantification in Energy Efficient Building Performance Simulations. Proceedings of the 2^nd International High Performance Buildings Conference, Purdue, IN, 2012.