Hybrid Probabilistic Search Methods for Simulation Optimization

Discrete-event simulation based optimization is the process of finding the optimum design of a stochastic system when the performance measure(s) could only be estimated via simulation. Randomness in simulation outputs often challenges the correct selection of the optimum. We propose an algorithm that merges Ranking and Selection procedures with a large class of random search methods for continuous simulation optimization problems. Under a mild assumption, we prove the convergence of the algorithm in probability to a global optimum. The new algorithm addresses the noise in simulation outputs while benefits the proven efficiency of random search methods.