Techniques for optimization via simulation: an experimental study on an (s, S) inventory system

Various methods have been proposed to conduct optimization via discrete-event simulation. Prevalent among these are gradient-based algorithms, and more recently, so-called retrospective approaches that deterministically optimize over simulated outcomes. In this paper we compare and contrast the performance of these two approaches for the optimization of a periodic review (s S) inventory system. In practice, gradient-based methods often exhibit undesirably slow convergence, whereas the retrospective approach can be computationally burdensome. We propose a hybrid algorithm that combines aspects of both approaches to mitigate these shortcomings.