Stochastic trust region response surface convergent method for continuous simulation optimization
Abstract
Response surface methodology (RSM) is a widely-used method for simulation optimization. Its strategy is to explore small subregions of the parameter space in succession instead of attempting to explore the entire parameter space (often complicated) directly. This method is particularly suitable to complex systems where little prior knowledge is available. While RSM is popular in practice, it has two well-known shortcomings. Firstly, it is not automated; human involvement is required in each step of the search process. Secondly, RSM is a heuristic procedure without convergence guarantee; the quality of the final solution in RSM cannot be quantified. We propose Stochastic Trust Region Response Surface Convergent Method, abbreviated as STRONG, that stresses these two problems. STRONG combines classic RSM with trust region method developed for deterministic optimization to eliminate the requirement of human intervention and to achieve the desired convergence property. The white-noise assumption in classic RSM is also relaxed to heterogeneous normal errors, general non-normal errors, and non-additive errors. These relaxations can extend the applicability of the proposed method to a wider problem spectrum. Since STRONG is a response-surface-based framework, it allows for incorporation of many powerful statistical tools such as design of experiments and regression techniques. The computational advantage is increasing with the size of problem when appropriate experimental designs (including screening methods) are employed.
Degree
Ph.D.
Advisors
Wan, Purdue University.
Subject Area
Statistics|Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.