Selecting the optimum by searching and ranking procedures in simulation-based optimization

Given a set of competing system alternatives to be evaluated and compared via simulation, Ranking and Selection (R&S) procedures are commonly applied to select the best system with respect to a predefined performance measure. In this paper we focus on two major classes of R&S techniques usually referred to as the subset selection and indifference-zone formulations. In particular, we discuss the performance of primitive and combined procedures that, at every iteration, evaluate different system configurations by sampling multiple or single additional simulation output observations to deal with complex systems. Procedure application is presented for different test cases in which either a small number of system configurations are known a priori or a large number of configurations are actually generated during simulation run by means of simulation-based optimization algorithms. Preliminary numerical results are given with reference to performance measures within the sub-systems of a realworld complex logistic system.