A computational comparison of simulation optimization methods using single observations within a shrinking ball on noisy black-box functions with mixed integer and continuous domains

摘要

We focus on simulation optimization algorithms that are designed to accommodate noisy black-box functions on mixed integer/continuous domains. There are several approaches used to account for noise which include aggregating multiple function replications from sample points and a newer method of aggregating single replications within a “shrinking ball.” We examine a range of algorithms, including, simulated annealing, interacting particle, covariance-matrix adaption evolutionary strategy, and particle swarm optimization to compare the effectiveness in generating optimal solutions using averaged function replications versus a shrinking ball approximation. We explore problems in mixed integer/continuous domains. Six test functions are examined with 10 and 20 dimensions, with integer restrictions enforced on 0%, 50%, and 100% of the dimensions, and with noise ranging from 10% to 20% of function output. This study demonstrates the relative effectiveness of using the shrinking ball approach, demonstrating that its use typically enhances solver performance for the tested optimization methods.