Rapid Screening Procedures for Zero-One Optimization via Simulation

Some existing simulation optimization algorithms (e.g., adaptive random search) become pure random search methods and thus are ineffective for the zero-one optimization via simulation problem. In this paper, we present highly efficient rapid screening procedures for solving the zero-one optimization via simulation problem. Three approaches adopting different sampling rules and providing different statistical guarantees are described. We also propose efficient neighborhood search methods and a simple algorithm for generation of initial solutions, all of which can be incorporated into our rapid screening procedures. The proposed procedures are more adaptive than ordinary ranking and selection procedures because in each iteration they can eliminate inferior solutions and intelligently sample promising solutions from the neighborhood of the survivors. Empirical studies are performed to compare the efficiency of the proposed procedures with other existing ones.