Research On Modeling And Optimization Of Hybrid Parametric Experiment Design Based On Double Nested LS-SVR

There are many process parameters that affect product quality characteristics.In addition to conventional scalar parameters,functional parameters with obvious functional characteristics have also become the object of process parameter research.Functional parameters are different from scalar parameters.Its effect on quality characteristics is a continuous,dynamic process that changes with time.The effect on product quality is also produced by the continuous change of the curve.For manufacturing processes that contain functional and scalar hybrid parameters,complex parameter interactions,and complex parameters that have a complex effect on quality characteristics,how to optimize the parameters through experimental design to achieve parameter optimization has become an issue to be studied.To this end,a hybrid parametric experimental design modeling and optimization method based on double-layer nested LS-SVR is proposed,as follows:Firstly,the number of control points and the feasible region are determined by analyzing the prior curve form of the functional parameter,and the weighted summation of the Bernstein polynomial basis function and the control point is used to express the functional parameter in the form of a Bezier curve.The Bezier curve is a widely used approximation curve,and its description is relatively simple and convenient.As long as the control points are given,the shape of the curve can be determined.As the control points change,the curve shape also changes.Control points can describe complex curve shapes,so Bezier curves can be used to implement self-modeling of functional parameters,building an inner model;Secondly,the control points of the Bezier curve are used as the representative of the functional parameters,and the functional parameters are fused with the scalar parameters through the hyper-Latin square experiment to obtain the sample points.Latin Hypercube Sampling is a type of space-filling design,and its essence is a stratified sampling.The experimental point arrangement is more flexible,the required sample size is smaller,the sample point distribution is more uniform,it is not easy to cause dimensional disaster,and it is more suitable for the complex interaction process of hybrid parameters.Furthermore,the least squares support vector regression(LS-SVR)suitable for small samples is selected as the basic method for hybrid parameter modeling.Because the external appearance of functional parameters is a curve,in order to accurately calculate the distance between the curves,the kernel function of LS-SVR is improved by using semi-metrics,and the distance between the functional parameters and the scalar parameters is The distance is separated and measured to improve the LS-SVR model,building a hybrid parameter outer layer model,Together with the inner model to form a double-layer nested model.Finally,the improved kernel function is used to model the hybrid parameters and the genetic algorithm is used to optimize the model,so as to obtain the desired optimized parameter combination and achieve parameter optimization.Simulation and empirical studies show that: functional parameter design based on the prior curve form can help obtain representative sample points,and the model performs better after modeling;using semi-metrics to embed functional parameters from the "functional" perspective as a whole Parametric modeling is feasible and effective.