Kriging-grnn High Order Mixed Response Surface Model Construction Method And Simulation Application

With the development of high efficiency and intelligence of mechanical and electrical products,the simulation model of mechanical and electrical products is becoming more and more complex.In order to improve the quality of mechanical and electrical products,at the same time,the simulation model is required to have high accuracy,convenient to accurately predict the performance of mechanical and electrical products,and improve the quality of mechanical and electrical products.Theoretically,high-order response surfaces can represent complex simulation models,such as polynomial response surface,Kriging,gradient enhanced Kriging,radial basis function,support vector,generalized regression neural network(GRNN)and so on.However,the high-order response bread contains a large number of undetermined parameters.In order to accurately adjust the parameters,a large number of experimental data are needed for calculation,and a large number of experimental data calculation will lead to high calculation costs.In addition,there are usually fluctuation deviations in the construction of response surface models by these methods,and a single method will generally lead to the problem of overfitting and underfitting.In order to solve the problem of fluctuation deviations of response surface algorithm,the traditional processing methods usually use a large number of data to adjust the parameters in the response surface,which will not only lead to over-fitting,but also increase the calculation time needed to build the response surface.In order to solve this problem,on the basis of the existing response surface theory research,the fluctuation deviation is normalized.Based on GRNN and Kriging interpolation algorithm,this paper proposes Kriging-GRNN High Order mixed response surface model construction method and simulation application(KGRNN).The main research contents and innovations of this paper are as follows:(1)Through the research and comparison of ANN,RBF,Kriging,GRNN algorithms,ANN algorithm is easy to fall into the local optimal solution.RBF is easy to produce the wrong heap direction,and the fitting results tend to simple surfaces.Kriging was closely related to regression analysis.A large amount of data is needed,and GRNN has strong fault-tolerance and robustness,which is suitable for solving nonlinear problems.KGRNN algorithm uses Gaussian kernel function,adjusts the parameters less,speeds up the computational efficiency of the response surface model.(2)Compared with the traditional single response surface algorithm,KGRNN is a hybrid response surface algorithm of Kriging and GRNN,GRNN algorithm fits the macro characteristics of mechanical and electrical products,and Gaussian function is used in the pattern recognition layer of GRNN algorithm.Gaussian function is used to map the parameters of mechanical and electrical products to high dimensions,and different data sets are used for iterative training.There is some fluctuation deviation in GRNN algorithm.According to the form of fluctuation deviation,the fluctuation deviation is normalized.Kriging algorithm fits the microscopic characteristics of the fluctuation deviations,and the normalized fluctuation deviations is substituted into the regression Kriging algorithm.The multivariate normal distribution form of the fluctuation deviations is fitted,and the fluctuation deviation solved by the regression Kriging algorithm is subtracted from the GRNN.The accuracy of GRNN algorithm is improved.(3)Through simulation experimental,the response surface algorithms of KGRNN,KRBF,GRNN,Kriging and RBF are compared.Compared with the traditional algorithm,the accuracy of KGRNN algorithm is improved by 0.1 to 10 times,and the parameters adjusted by KGRNN algorithm are less.Compared with GRNN,KRBF and RBF algorithms,the speed of KGRNN response surface model construction is increased by 3 to 5 times.The robustness and reliability of KGRNN algorithm are also proved by the engineering examples of composite square tube and cylindrical stiffened fuselage structure.