MULTISTEP INPUT REDUCTION FOR HIGH DIMENSIONAL UNCERTAINTY QUANTIFICATION IN RAVEN CODE

摘要

This manuscript focuses on the efficient construction of surrogates that can be employed in place of complex systems. The surrogates under consideration are fast to evaluate and yet accurate, which take the form of generalized polynomials chaos. Here, we utilize the principal component analysis (PCA) to reduce the number of input parameters, and we employ adaptive stochastic collocation (ASC) method to reduce the number of functions evaluations. Under the PCA and ASC methods, we propose a method that allow us to construct a surrogate that can effectively handle large number of input parameters and even more when adaptive high-dimensional model reduction (AHDMR) algorithm are employed to further reduce the dimensionality of input space. In the past year, this approach has been developed and implemented in RAVEN.