| Summary: | 碩士 === 國立中興大學 === 機械工程學系所 === 97 === Abstract
This thesis incorporates data mining into optimization process. The useful information embedded in the data can be dug out to enhance the computational efficiency and produce better solutions. For unconstrained optimization problems, the data mining is used to find the design space that might contain the global solution. For constrained optimization problems, it is used to find the possible feasible regions. The Sequential quadratic programming (SQP) is then used to find the optimum solution in the identified areas to see whether the chance to find the global solution is increased. Based on test results, it shows that for high-noised multi-modal problems, data mining and SQP may not be able to find the global solution. Therefore, the evolutionary algorithm will be incorporated into the optimization process in the second part of this thesis.
The evolutionary algorithm searches the design space using multiple points simultaneously. If the design space can be reduced, then the computational time spent will be reduced and the chance to find the global solution will be increased. In order to save the computational time for structural optimization problems, the artificial neural network is employed to get the approximate results of structural analyses.
This thesis uses evolution strategy to search for the optimum solution in design space found by data mining. For structural optimization problems, neural networks are used to replace exact finite element analyses. The SQP is used in the last step to search for the exact optimum solution. Several test problems show that the proposed approach not only can find better solutions but also spends less computational time.
|
|---|
