Fuzzy controlled simulation optimization

Simulation optimization deals with finding the values of input parameters of a complex simulated system which result in desired output. Traditional techniques may require an enormous amount of simulation runs to evaluate the system. To alleviate this problem, the proposed work provides the means of incorporating knowledge, expressed in natural language, that is often available among analysts and decision makers. Using convenient linguistic representations, the proposed mechanism can satisfy, vaguely stated goals to a high degree (e.g. "high utilization" or "low inventory"). This mechanism is also able to generate an approximate Pareto optimal set in the presence of multiple goals. The optimization strategy used here depends on a fuzzy controller guided by a set of rules derived from statistical concepts, response surface models, and experts' knowledge, To illustrate this approach we present computational experiments on the design of a flow line manufacturing system (in terms of a tandem of queues with blocking) with one and two goals. The actions derived from the controller, using approximate reasoning, are able to generate a high quality solution in only a few iterations, The results are compared extensively with those obtained with a genetic-based approach. (C) 2002 Elsevier Science B.V. All rights reserved.