Paper Conference

Proceedings of BSO Conference 2012: 1st Conference of IBPSA-England


Uncertainty-Weighted Meta-Model Optimization in Building Energy Models

Bryan Eisenhower, Vladimir Fonoberov, Igor Mezic´

Abstract: In this paper we discuss a method to efficiently optimize whole-building energy models which contain a large number of candidate optimization variables and have discontinuous cost surfaces with multiple minima. The approach leverages uncertainty propagation and sensitivity analysis to identify critical parameters that are most effective for optimization. Large discontinuities in the cost function are identified using a filtering method and reduced order meta-models are created in partitioned subsets of the global feasible set that are separated by these discontinuities. A derivative-free optimization algorithm is used while employing an uncertainty-weighted cost function to obtain the best optimized solution for each feasible subset. This method is tested on a large office building modeled in EnergyPlus but can be adapted to other models and modeling software.
Pages: 95 - 101