Space partitioning in piecewise metamodeling: a graphical approach

This paper presents a graphically-based space-partitioning methodology for piecewise metamodel building. The method is based on using data displays—in particular, the so-called ordinal plots—leading to a metamodel with acceptable prediction accuracy on a piece-by-piece basis, not just on average. The proposed methodology is general and can be applied to various metamodel types. A one-dimensional example is first used to demonstrate the approach, then the methodology is tested on a number of analytical examples in higher dimensions, one of which corresponds to an electronic filter.