Combined pattern search and ranking and selection for simulation optimization

A new algorithm class is presented for optimization of stochastic simulation models. The algorithms, which combine generalized pattern search (GPS) with ranking and selection (R&S), require "black-box" simulation evaluations and are applicable to problems with mixed variables (continuous, discrete numeric, and categorical). Implementation of the mixed-variable generalized pattern search with ranking and selection (MGPS-RS) algorithm with three different R&S procedures is demonstrated and tested on a small set of standard test functions. Results of this preliminary performance evaluation are summarized and compared with existing search methods.