Simulation Optimization in Uncertain Environments: An Evolutionary Approach

Abstract

The design of manufacturing systems often involves simulation optimization approaches to search for the best system performance. Metaheuristic approaches are more and more used to optimize simulation models. In most existing approaches, the optimization is performed with fixed environmental conditions (e. g., part demand, breakdown rates are assumed to be perfectly known). However, in practice the actual data about the system may differ from those used in the simulation model (e. g., modification of the part mix). To cope with this, the candidate solutions, in a simulation optimization process, can be compared on various possible environments, using such principles as those proposed by Taguchi. Unfortunately, when metaheuristics are used, simulation optimization can be very much time consuming, since each solution has to be compared on a number of different environments. In order to provide robust solutions in a more reasonable time, we propose a two stage heuristic search. First, a restricted set of n promising solutions is identified using an evolutionary multimodal simulation optimization process, using the concept of base environmental scenario, recently published. Then robustness evaluation on many environments is performed only on these n promising solutions and the most robust can be chosen. This approach is illustrated on a supply chain problem where several parameters have to be defined. As a result, 13 solutions are found. The most robust solution is not the one that yields the best results in the environment assumed by the decision maker. This result shows how important it is to be able to consider several environments in the simulation optimization process.