A Novel Search Algorithm Based on Taguchi Method for Discrete Global Optimizations

• Main Authors: Tao-Wei Chen, 陳道偉
• Other Authors: Hsin-Chuan Kuo
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/46165686305817479898

碩士 === 國立臺灣海洋大學 === 系統工程暨造船學系 === 97 === Taguchi Method applied in Design of Experiments has earned much credits of getting a feasible design setting economically and effectively. However, as a fraction factor design of experiments, Taguchi method works as only a local optimization solver. Further as the complexity of a problem comes higher, to get a reliable solution via Taguchi method needs more work than expected, and sometime the solution obtained even goes to an unreliable one. To copy with the difficulty and improve the performance of Taguchi method, the author proposed two creative methods under the framework of Taguchi method. They are sequential Taguchi method and neighborhood Taguchi method. The proposed two methods were applied in the work to three uni-modal benchmark problems and then verified as effective solvers, able to obtain solution closer to the exact solution than the original Taguchi method. Moreover, the two improved methods works as global optimizer, on more than local optimizer as the Taguchi method. Base on the neighborhood Taguchi method, the better one of the two above-proposed methods, in the thesis, the eliminated-type and cooperative-type Taguchi methods are proposed to deal with the design of experiments having extra number of levels and number of factors larger than those of any one orthogonal array. Through the above-studied benchmark problems, the performance of two proposed approaches are verfied. Finally, the proposed methods were applied to a system engineering problem for determining its reliability in maximization. In collaboration with the Fuzzy theory, the Weighting approach and the Grey relational analysis, the proposed methods are further extended in the thesis to the work for solving the multi-objective problems.