Using data mining technique to search partial multi-objective optimal solutions in some designated objective ranges

• Main Authors: Pin-Hao Wu, 吳品豪
• Other Authors: Ting-Yu Chen
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/45843117282731577363

碩士 === 國立中興大學 === 機械工程學系所 === 102 === Data mining is applied to solve multi-objective optimization problems in this thesis. In order to get rules from data mining, we put some sample points in the design space using uniform design method. The number of sample points is determined by the number of variables and the complexity of the functions. The values of objective functions of the sample points are then computed. Based on user’s demand, some specific objective intervals are selected. The classification and clustering techniques in data mining are used to find the ranges of design variables that may generate objective values in the selected intervals. To increase the accuracy of the ranges found, a second stage of classification and clustering is performed on the ranges found previously. Within the ranges found, one point is generated randomly as the initial point for solving multi-objective optimization problems. The sequential quadratic programming (SQP) is incorporated with the weighted sum method or compromise programming method to search the pareto-optimal solutions, and the solutions are expected to be in the selected objective intervals. The solutions obtained will be compared with the complete pareto fronts in related papers. Using the method proposed in this thesis, there is no need to find the complete pareto front. Only interested pareto solutions are found. This not only saves a lot of computational time but also satisfies the user’s need.