The Application Of Improved Immune Evolutionary Algorithm On Multi Kernel SVM Parameters Optimization

Support vector machine(SVM)is a product of the development of statistical learning,based on the training error as constraint condition of optimization,with the target of the minimum confidence limit value,based on structural risk minimization criterion,and it integrates the maximum margin hyperplane,Mercer kernel function,convex quadratic programming and the nonnegative relaxation factor technologies such as learning methods.SVM algorithm has already replaced the neural network algorithm because of the fast convergence,stability and strong generalization ability and other characteristics.In the process of the study of feature vector,when low dimensional feature space can not separated by a line,with the introduction of kernel function,the low dimensional inseparable vector is mapped to high-dimensional feature space,in the high-dimensional space structure decision function,thus will originally inseparable data into separable data.The introduction of kernel function is to increase the classification ability of the learning phase.Due to the impact of the global parameters of the kernel function,local kernel function and kernel function,their classification performance is not very satisfactory.In addition,in the actual application,also it is difficult to use a single kernel function to express complex rules of various data hidden in practice.In recent years,scholars have put forward a model of multi-kernel SVM,it is a kind of flexibility is more wider application of the learning model based on kernel,and it now has become a hot research field of machine learning.To use a multi-kernel function SVM model can get more than a single kernel function SVM algorithm better classification performance.This paper proposes an optimization algorithm based on Multi-subgroup competition parallel immune evolutionary programming and multi-kernel SVM algorithm,and its application in the field of fault diagnosis and chaotic prediction.This paper describes the process of development and research status.of the SVM model.Proposed advantages and applications of SVM,as well as domestic and foreign research status of their application.SVM improves the traditional neural network algorithm in the face of the shortage of the condition of small samples,solve the problem of the traditional method can't avoid the local extremum and dimension disaster problem,and has a good classification performance.Secondly,it introduces the support vector machine algorithm based on multi-kernelfunction.In the multi-kernel function SVM,multi-kernel functions describe the data characteristics stronger,and kernel function has more ability to promote and enhance the decision-making function interpretability.In the popular intelligent optimization algorithm,this paper based on immune evolutionary algorithm as the idea and proposes an optimization algorithm based on MCPIEP algorithm.Through the widely used five benchmark functions,the contrast experiment shows that the optimization algorithm can improve ability and stability of the decision function,more ability to promote and enhance the decision-making function interpretability.Finally,the algorithm is applied to the fault detection and chaos prediction field,and it achieved good effect verified by the experiment.