During the last 1-2 decades, simulation optimisation of discrete event dynamic systems (DEDS) has made considerable theoretical progress with respect to computational efficiency. The score-function (SF) method and the infinitesimal perturbation analysis (IPA) are two candidates belonging to this new class of methods, where one single simulation run in principle is sufficient for the estimation of any desired number of partial gradients. Embedded in an iterative set-up both the SF and the IPA methods belong to the class of Stochastic Approximation (SA) algorithms and furthermore if the gradients are unbiased, the SA-algorithm will be known as a Robbins-Monro-algorithm. The present work will focus on the SF method and show how to migrate it to general types of discrete event simulation systems, in this case represented by SIMNET II, and discuss how the optimisation of the functioning of a Job-Shop can be handled by the SF method.
Logistics Changes in the New Century, 2000, p. 31-46
Simulation; Optimisation; Score Function; Stochastic Approximation; Robbins-Monro; Job-Shop; SIMNET II
Main Research Area:
XIIth NOFOMA conference - NOFOMA 2000
Aarhus School of Business, Department of Management Science and Logistics