A genetic algorithm for robust simulation optimization
An automated, robust optimization algorithm coupled with simulation would provide a powerful tool for the analysis and design of systems. Such a technique would function on a wide variety of problems and find optimal or near-optimal solutions in the least amount of time possible. The technique would search the problem's solution space by selecting system inputs based on simulation output. The purpose of this thesis was to investigate the feasibility of using a genetic algorithm for robust simulation optimization. A genetic algorithm was developed from literature consultation and two experiments. It was applied to 20 simulation test problems. To measure the algorithm's robustness, its parameters were not fine-tuned for the different problems. The test problems were developed to represent a wide variety of realistic problems of various sizes and characteristics. The algorithm proved to be robust to all the test problems because it was able to handle all of their different characteristics. The genetic algorithm found a solution as good as a gradient technique's solution on a benchmark test problem taken from literature, better solutions than a simulated annealing algorithm's solutions on 85% of the 20 test problems, and significantly better solutions than simulated annealing on 45% of the problems. The genetic algorithm out-performed simulated annealing by a larger margin on large problems, but required more replications on most of the problems. From the results, this genetic algorithm was shown to be a feasible and promising candidate for robust simulation optimization.
School Location:USA - Ohio
Source Type:Master's Thesis
Keywords:genetic algorithm s robustness gradient technique
Date of Publication:01/01/1996