Variable Reduction for Surrogate Modelling

In this paper, we present a three-step procedure for the reduction of independent variables u for surrogate modelling. First, the linear material balances are introduced to reduce the number of surrogate models which need to be fit. Second, partial least square (PLS) regression of a sampled space is performed to obtain new variables (components) and third, the new components are used as input variables for the fitting of a nonlinear surrogate model. The application of PLS reduces the number of independent variables through the introduction of linear combinations of the original independent variables u. The proposed procedure is applied to two examples, the first describes a simple pipe model in which the minimum number of new independent variables u′ is known and which hence serves as a proof of concept. The second examples considers the reaction section of the ammonia synthesis gas loop for integrated submodels. In both examples, it is possible to reduce the number of independent variables by at least a factor of 2 while maintaining accuracy.