The OPTLSO Procedure

The OPTLSO procedure uses a hybrid combination of genetic algorithms (Goldberg, 1989; Holland, 1975), which optimize integer and continuous variables, and generating set search (Kolda, Lewis, and Torczon, 2003), which performs local search on the continuous variables to improve the robustness of both algorithms. Both genetic algorithms (GAs) and the generating set search (GSS) have proven to be effective algorithms for many classes of derivative-free optimization problems. When only continuous variables are present, a GA usually requires more function evaluations than a GSS to converge to a minimum. This is partly due to the GA’s need to simultaneously perform a global search of the solution space. In a hybrid setting, the requirement for the GA to find accurate local minima can be relaxed, and internal parameters can be tuned toward finding promising starting points for the GSS.