Mixed Parameter Modeling And Optimization Based On Support Vector Regression

Manufacturing is the pillar industry of the country,with the advancement of technology and the diversification of products,a large number of complex interactions have emerged in the manufacturing process,which are characterized by numerous parameters,nonlinear correlation between parameters and quality characteristics,and input parameters not only contain scalar parameters,but also have functional parameters.The function parameter is a variable that is different from a traditional scalar parameter.This parameter does not exist in the form of "points" but in the form of "functions(curves)".Its optimization goal is to determine a suitable curve equation rather than a certain level of value.Design of experiments is one of the main methods for parameter optimization in the production process.Its main purpose is to analyze which input parameters significantly affect the output quality characteristics and how to determine the values of the input parameters so that their quality characteristics are optimal.However,the traditional parameter optimization theory and method based on experimental design only aim in the case that the parameters and quality characteristics are all scalars,not study the process of complex interaction with functional parameters.Based on this,the paper studies the parameter optimization of complex interactive process from the perspective of functional parameters,and proposes a hybrid regression modeling and optimization based on the SVR,which contains functional parameters and scalar parameters.The specific research contents are as follows:(1)Aiming at the problem of modeling functional parameters,the B spline curve is introduced.The B spline curve is piecewise polynomial curve.It can reduce the order of fitting curve,and has a good fitting effect for complex curve.Therefore,the B spline curve can be used to model functional parameters.First of all,a set of data points is randomly obtained by using the super Latin square design,and then the control points of the B spline curve are obtained.Then,the B spline basis function is obtained by selecting the proper number of B spline curves and the number of nodes.According to the expression of the B spline curve,the functional parameter modeling is realized.(2)The support vector regression machine(SVR)is introduced to model the complex interaction process modeling problem with functional parameters and scalar parameters.Support vector regression(SVR)is generally accepted as the best universal machine learning theory for small samples,a better predictive model and an optimization model can be established with a small sample size.First of all,we select appropriate experimental design methods based on the established functional and scalar parameters,such as full factorial design with center point or uniform design,and get the output quality characteristic value,so as to get the training set needed for modeling.According to the training set,select appropriate support vector regression parameters,build support vector regression machine hybrid model,then select the appropriate evaluation index to evaluate the model,and finally,the genetic algorithm is used to optimize the model,the better input parameters are obtained,and the desired quality characteristic values are obtained.(3)Finally,the method of the paper is applied to the complex simulation function and the optimal case of polymer injection process,simulation and case studies show that under the condition of small sampling and small sample size,the proposed method can better establish a mixed regression model with functional parameters and scalar parameters,which can effectively reduce the quality improvement cost and achieve the improvement of product quality.