Combining Projection Pursuit and Hybrid Subspace Methods for Reduced Order Modeling

摘要

To help reduce the computational cost required to build nonlinear surrogate models, we adopt a two-step approach. In the first step, the effective number of parameters is reduced via a hybrid subspace method developed previously [1]. In the second step, a surrogate model is constructed using a projection pursuit algorithm, which allows one to recast the problem of finding a surrogate into a minimization problem. This allows one to identify the most influential components of the nonlinear models with reduced number of forward model executions.