Supersaturated experimental designs. New approaches to building and using it: Part I. Building optimal supersaturated designs by means of evolutionary algorithms

A new procedure for the building and selection of supersaturated design matrices is presented. The procedure is useful in generating screening experimental designs in the range 8–22 runs. An evolutionary algorithm is used to select between all possible candidate columns, which in turn, are a function of the selected run number, those producing the optimal matrix. Optimality, as defined by three sequentially applied common criteria (Es2, n0, m0), is used as fitness functions in the evolution algorithm. The problem in the construction of an optimal design matrix as a particular subset of a much larger universal set of potential solutions needs specially problem-adapted genetic operators. Several have been tested and applied. To make the procedure practical, a toolkit has been developed which allows, in a reasonable computation time, to build and select well characterised experimental supersaturated designs for a given run and factor numbers.