| Summary: | 博士 === 國立清華大學 === 化學工程學系 === 91 === Combinatorial optimization problem have become more and more important in many areas of chemistry and chemical engineering research. The optimum solution of those problems usually is solved via the heuristic search method, such as genetic algorithm, simulated annealing method, evolution algorithm etc. Although the heuristic methods are known to be powerful in a large number of papers, they unfortunately required extensive computation time since the number of search iteration is enlarged to escape the trap of local optima. A high computation cost or expensive experiment validation usually is a critical issue, particularly in the case of many replications or when a long run length in simulation is required. For addressing the limitation of existing problem such as a high computation, expensive experiment validation, we suggested that a simple prediction model using a priori available data can be constructed using a generalized regression neural network. An index of our uncertainty about a point in the search space can also be established using information entropy. An information free energy combined the two indices to direct the search so that importance sampling is performed. In the thesis, the proposal method to be tested through a few different type of benchmark problems, such as RPV problem and NK problem, being used to model the optimization problem involved in combinatorial synthesis and library design, heat exchanger network synthesis problem in chemical engineering process and cutting stock problem encounted in the paper-converting industry, steel industry and glass industry. We showed that when importance sampling is performed, the combinatorial technique becomes much more effective. The improvement in efficiency over undirected method is especially significant when the size of the problem become very large.
|
|---|
