|
| Design space reduction based on the metamodeling and clustering method |
| Received:May 26, 2010 Revised:August 18, 2011 |
| View Full Text View/Add Comment Download reader |
| DOI：10.7511/jslx20122016 |
| KeyWord:metamodel surrogate model Kriging model Clustering method |
| Author | Institution |
| 周仕明 |
国防科学技术大学 指挥军官基础教育学院, 长沙 |
| 李道奎 |
1. 国防科学技术大学 指挥军官基础教育学院, 长沙 |
| 唐国金 |
国防科学技术大学 航天与材料工程学院, 长沙 |
|
| Hits: 2625 |
| Download times: 1489 |
| Abstract: |
| A design space reduction methodology was proposed, which based on metamodeling and clustering method to reduce the design space to a relatively small region. This methodology is composed of three main steps. In the first step, the metamodel is constructed according to the initial samples. In the second step, new samples are generated, the function and derivative of the sample are computed based on the metamodel. The central and number of cluster is determined by the Nearest Neighbor method. The new samples are partitioned into several clusters by Fuzzy C-mean method and design subspaces are generated. Finally, several subspace filters out due to the function mean of every subspace. The global optimization result of objective function is obtained from the residual design subspaces. The test problem and engineering instance show the accuracy and efficiency of proposed method. |
|
|
|