Kara kutu benzetim modellerini eniyileyen yöntemlerin nümerik karşılaştırılmaları: RSM ve SPSA

By the developments in computer technologies, there has been a significant increase on academic studies and practical applications about simulation optimization in the last decades. The major advantage of this augmentation, which also positively affects commercial software, is the opportunity of examiner a large number of design solutions and to obtain the values of various performance measures compared to the traditional methods.Simulation optimization methods can be classified as white box methods and black box methods. While white box methods enabling the practitioners to estimate the random responses and their gradient vectors, black box methods use the simulation outputs of the random responses to approximate their gradient vectors. Although white box methods outperform black box methods, black box methods shall be preferred in a case which the knowledge of the stochastic structure of the system is not available to the practitioners.In this study, performances of black box methods Response Surface Methodology and Simultaneous Perturbation Stochastic Approximation are compared numerically considering that the computer budget is limited and each simulation run is time-consuming. Comparisons are realized on the newsvendor problem, (s, S) inventory model and generic call center problem.The computer program of RSM and SPSA algorithms are coded in Matlab 7.6 for the newsvendor problem and (s, S) inventory model. The call center model is first designed in Arena, then RSM and SPSA steps are applied manually based on the results get by this model.