informs ® doi 10.1287/opre.1050.0237 © 2006 INFORMS We propose an optimization-via-simulation algorithm, called COMPASS, for use when the performance measure is estimated via a stochastic, discrete-event simulation, and the decision variables are integer ordered. We prove that COMPASS converges to the set of local optimal solutions with probability 1 for both terminating and steady-state simulation, and for both fully constrained problems and partially constrained or unconstrained problems under mild conditions.

ABSTRACT: Optimization techniques have been used to search for optimal values for decision variables and input variables associated with a simulation. More recently we have explored a mixed-method approach, mixing optimization and code modification, for “coercing ” simulations to meet new requirements. The coercion process, P, transforms a simulation that meets requirement R to meet a new requirement, R', without necessarily resorting to redesign or reimplementation. Coercion is a semi-automated process because it combines optimizations and code modifications. A coercion can be characterized as a regular expression P = [ m*, o *] * where m and o represent code modification and optimization respectively. P can consist of any number, including zero, of modifications and/or optimizations in any order. The use of optimization is encouraged because it increases the level of automation in the coercion process. We explore the question of determining best optimization techniques for coercion. We begin by considering classifications of optimization techniques already existing in the simulation community and apply these in the context of coercion. We consider issues such as computation time, set-up time, and avoidance of local minima that contribute towards the ease of use of each optimization technique. We discuss insights we expect to gain from the use of optimization and identify tools necessary to gain these insights. Additionally, we discuss how the result of one optimization may affect the outcome of another in a coercion sequence, P. We expect our results to serve as a guide to optimization method selection for simulation practitioners who wish to employ coercion. 1.

We present a functional framework for automated Bayesian and robust mechanism design based on a two-stage game model of strategic interaction between the designer and the mechanism participants, and apply it to several classes of two-player infinite games of incomplete information. Our approach yields optimal or nearly optimal mechanisms in three application domains using various objective functions. By comparing our results with known optimal mechanisms, and in some cases improving on the best known mechanisms, we show that ours is a promising approach to parametric design of indirect mechanisms. 1

Based on a discrete-event simulation model, Simulationbased Real-time Decision-Making (SRDM) is an innovative approach to real-time, goal-directed decision-making. When applied to a flexible manufacturing system, SRDM makes better decisions than most fixed policies, such as deterministic, stochastic and manual. SRDM even improved over fixed policies optimized within a class of policies by OptQuest, in our numerical experiments. Compared to these fixed policies, SRDM shows greater improvement for more complex systems and is quite robust with respect to modeling errors. SRDM provides an improvement over fixed policies by its ability to implement adaptive policies. Since most real-time decisions in currently deployed manufacturing systems are made either manually or by using fixed policies, our results suggest that using SRDM instead could lead to significant improvement in operating performance.

Metaheuristics have been established as one of the most practical approach to simulation optimization. However, these methods are generally designed for combinatorial optimization, and their implementations do not always adequately account for the presence of simulation noise. Research in simulation optimization, on the other hand, has focused on convergent algorithms, giving rise to the impression of a gap between research and practice. This chapter surveys the use of metaheuristics for simulation optimization, focusing on work bridging the current gap between the practical use of such methods and research, and points out some promising directions for research in this area. The main emphasis is on two issues: accounting for simulation noise in the implementation of metaheuristics, and convergence analysis of metaheuristics that is both rigorous and of practical value. To illustrate the key points, three metaheuristics are discussed in some detail and used for examples throughout, namely genetic algorithms, tabu search, and the nested partitions method.

We consider coordination among stocking locations through replenishment strategies that explicitly take into account lateral transshipments, i.e., transfer of a product among locations at the same echelon level. Our basic contribution is the incorporation of supply capacity into the traditional transshipment model. Our goal is to analyze the impact on system behavior and on stocking locations ’ performance of the fact that the supplier may fail to fulfill all the replenishment orders. We therefore formulate the capacitated supply scenario as a network flow problem embedded in a stochastic optimization problem. To solve this problem, we devise a sample path optimization method together with infinitesimal perturbation analysis (IPA) for computing gradients. We find that, depending on the production capacity, system behavior can vary drastically. Moreover, in a productioninventory system, we find evidence that either capacity flexibility (i.e., extra production) or transshipment flexibility (i.e., pooling) is required to maintain a desired level of service. Simulation applications: Sample Path Optimization; Infinitesimal Perturbation Analysis.

Simulation optimization today is an important branch in the field of heuristic optimization problems. Several simulators include built-in optimization and several companies have emerged that offer optimization strategies for different simulators. Often the optimization strategy is a secret and only sparse information is known about its inner workings. In this paper we want to demonstrate how the general and open optimization environment HeuristicLab in its latest version can be used to optimize simulation models.

Congested networks involve complex traffic dynamics that can be accurately captured with detailed simulation models. However, when performing optimization of such networks the use of simulators is limited due to their stochastic nature and their relatively high evaluation cost. This has lead to the use of general-purpose analytical metamodels, that are cheaper to evaluate and easier to integrate within a classical optimization framework, but do not capture the specificities of the underlying congested conditions. In this paper, we argue that to perform efficient optimization for congested networks it is important to develop analytical surrogates specifically tailored to the context at hand so that they capture the key components of congestion (e.g. its sources, its propagation, its impact) while achieving a good tradeoff between realism and tractability. To demonstrate this, we present a surrogate that provides a detailed description of congestion by capturing the main interactions between the different network components while preserving analytical tractable. In particular, we consider the optimization of vehicle traffic in an urban road network. The proposed surrogate model is an approximate queueing network model that resorts to finite capacity

Consider the context of selecting an optimal system from amongst a finite set of competing systems, based on a “stochastic ” objective function and subject to multiple “stochastic ” constraints. In this context, we characterize the asymptotically optimal sample allocation that maximizes the rate at which the probability of false selection tends to zero. Since the optimal allocation is the result of a concave maximization problem, its solution is particularly easy to obtain in contexts where the underlying distributions are known or can be assumed, e.g., normal, Bernoulli. We provide a consistent estimator for the optimal allocation, and a corresponding sequential algorithm that is fit for implementation. Various numerical examples demonstrate where and to what extent the proposed allocation differs from competing algorithms. 1.

Abstract not found