Finding
All of these results demonstrate the adaptability and superiority of the method proposed.
Paper
SVM-Based Nonparametric Dual Response Surface Methodology
Citations: 1
Full textAbstract
Current researches on nonparametric dual response surface methodology (NPDRSM) focus on impro- ving fitting performance,but lose sight of improving generalization performance of models.Thus NPDRSM does not work well for the quality optimization and robustness design under the constraints of real industrial proces- ses.A new NPRSM was presented,which collected data by using equal intervals space filling,and then fitted process response mean and variance models by support vector machines (SVM).Meanwhile,it optimized pa- rameters in SVM by comparing the upper bounds of generalized error of different SVM models.The experiments show that,with the same design of experiment,the average generalized error of SVM-based NPDRSM decreases by 31.0% compared with kernel based NPDRSM,and by 51.8% compared with artificial neural networks (ANN)-based NPDRSM;when the generalized errors are close,the average sample size of SVM-based NP- DRSM decreases by 35.0% compared with kernel-based NPDRSM,and by 48.6% compared with ANN-based NPDRSM.The generalization researches with different sampling manners show that the equal intervals space filling is an acceptable design method under the situations without prior knowledge about the processes.All of these results demonstrate the adaptability and superiority of the method proposed.
Authors
Che Jian-guo
Journal
Journal of Tianjin University