Research On Optimization Method Of Lattice Structure Design Based On K-Fold SVR Metamodel

The design cycle of lattice structure product based on physical model or numerical simulation model is very long,it hinders the iteration of lattice structure product updates.The Support Vector Regression(SVR)is a metamodel with powerful nonlinear and high-dimensional processing capabilities,strong predictive performance and small sample requirements.It has widely used in engineering product design to replace physical model or numerical simulation model.The prediction accuracy of the SVR metamodel can directly affects the design optimization results.How to improve the prediction accuracy of the SVR metamodel has always been a research hotspot.For example,the method of adding sample points sequentially and parameter optimization have improved the prediction accuracy of the SVR metamodel from sample point selection and parameter selection,but the error information at the existing sample points is not fully considered.In addition,there is uncertainty in the SVR metamodel constructed based on a limited number of sample points.How to quantify the estimated uncertainty of the SVR metamodel is the key to the application of the SVR metamodel to the lattice structure product design.In this paper,based on the above two points,the K-fold cross validation technique and the SVR metamodel are combined to improve the prediction accuracy.Considering the uncertainty of the K-SVR metamodel,a K-SVR metamodel design optimization method is proposed.Finally,the method is applied to lattice structure product examples.Firstly,the K-fold cross-validation technology was used to extract the error information of the SVR metamodel at the existing sample points and establish K sub-SVR metamodels.The sample points are sorted according to the size of the error and the weight of the sub-SVR metamodels were assigned to integrate all the sub-SVR metamodels to achieve the establishment of a K-SVR metamodel.Mathematical examples were used to verify the effectiveness and superiority of the K-SVR metamodel.At the same time,the influence of different number of subgroups on the K-SVR metamodel was analyzed.Secondly,the influence of the uncertainty of the K-SVR metamodel was considered,leave-one-out method was used to quantify the uncertainty of the K-SVR metamodel under the setted confidence level.And then the K-SVR metamodel design optimization method is proposed,mathematical examples shows the feasibility of the method.Finally,the good prospects and advantages of the application of the lattice structure were described,the process of the cell lattice structure selection was described in detail.The K-SVR metamodel design optimization method proposed in this paper is applied to the design optimization of the car A-pillar and car seat carrier part with the lattice structure,and the design goal was successfully completed.Simultaneously,the validity of the design optimization method based on K-SVR metamodel in lattice structure product design is verified.