A Sequential Sampling Method Based On Support Vector Regression

Instead of computationally expensive simulation programmes,metamodel based design method,which involves metamodel and sequential sampling method,has been widely used in various fields of modern engineering applications.Comparing with sequential sampling method,metamodel is just regard as a modeling implement in recently study while few researches take them as a union in consideration.Besides,information about space distribution or cross validation is common adopted in filling process.The following is done in this thesis.Firstly,a brand new sequential sampling method called SV-Voronoi is proposed in this paper.SV-Voronoi sequential sampling method digests information of support vector provided by the support vector regression.The first step of SV-Voronoi method is to patition the entire design space according to Voronoi diagram.After determing the region with largest predict error,new sample point will be generated at the location which is farthest from the sensitive point in Voronoi cell.It shows that the proposed SV-Voronoi method is simple and robust.Furtherly,cross validatuion and grid search is introduced to overcome the mismatch between the size of sample points and parameters.The common practice of support vector regression application is to identify a set of parameters at the beginning and ignore the influence of parameters on metamodel.Along with the process unceasingly thorough,the size and distribution of sample points have changed whereas fixed parameters may result in accumulative error.Adusting the parameters according to sample points can solve this problem effectively.Finally,SV-Voronoi sequential sampling method is applied to the research in interface of controllable-pitch propeller hub.The result demonstrtes that the proposed method can establish response models accurately in different working conditions,which comfirms the ability of SV-Voronoi in engineering application.