Prescriptive simulation: A heuristic approach
Abstract
In this thesis, a methodology for the integration of a general purpose heuristic optimization algorithm and a commercial computer simulation language is developed to enhance and extend the benefits of using computer simulation to solve complex business problems. A unique version of the standard Simulated Annealing algorithm is developed that eliminates the need to tune the algorithm for each problem, thus producing a general purpose optimization framework. Concepts from Statistical Process Control form the basis of the proposed algorithm. Along with the proposed methodology and algorithm development, significant emphasis is focused on the issues involved in using a random variable-based technology, computer simulation to supply values for a deterministic-based technology, heuristic optimization. Results obtained demonstrate that the proposed algorithm can outperform the standard Simulated Annealing algorithm in two thirds of the test cases in terms of average results obtained. The proposed algorithm provided a maximum of eight percent improvement over the standard Simulated Annealing algorithm on similar problems where an improvement was obtained. When the standard Simulated Annealing algorithm provided better results, they were a maximum of only three percent better than the proposed method. Validation of the algorithm on an expanded set of known test cases provided similar results. The significance of this research to industrial engineering is a methodology and algorithm that will allow an Operations Research analyst to heuristically optimize a computer simulation model according to a pre-defined output response function without resorting to the iterative trial and error process that is in common practice today. This methodology will enable the analyst to quickly obtain boundary values of an output response function for any system under study, allowing the analyst to spend scarce project time on analysis and decision making of boundary conditions rather than on trial and error iterations in an attempt to find these conditions.
Degree
Ph.D.
Advisors
Sparrow, Purdue University.
Subject Area
Industrial engineering|Operations research
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.