Response Surface Methodology at Irregular Grids Based on Voronoi Scheme with Neural Network Approximator

Abstract

Traditional response surface methodology is based on polynomial approximators and designs prepared according to statistical requirements. In those cases when response surface cannot be modelled with single polynomial plane, a neural network approximator should be used. These approximators require a different design identification procedure. The author proposes to apply well-known Voronoi scheme to construct iteratively a new design.