Improved Support Vector Machine Based On Particle Swarm Optimization Algorithm

With the development of science and technology and the advancement ofsociety.industrial process appears complicated features on multivariablecoupling,strongly non-linear system,uncertainty model.It restricts the productionand development of industrial seriously.How do we forecast and control acomplexity industrial process is becoming a hot debated in the scholars.Artificialintelligence algorithms develops rapidly recently due to wide applicationfield,modeling-easily,weak mechanism.This paper has made a thorough researchon both PSO(particle swarm optimization) algorithm and SVM(support vectormachine),this paper studies and analyses on the zymolysis of penicillin to feedsupplement and the problem of load prediction in electric power system.Firstly,PSO has simple calculation,low complexity,little definitizedparameter,but it can guarantee that the algorithmic models has highly finallyaccuracy convergence.secondly,through the analysis,because of the PSO easilyslump the local trap,has low space exploring ability and rapid convergencequality.so we need improve the model.we introduce into exponential decreasinginertia weight,convergence factor and simulated annealing to improve theoriginality PSO.finally,through four groups of standard test functions to validatedthe improved PSO.The result reveal that the PSO of the Tabu search annealingprinciple has obvious advance in arithmetic speed and search capability.Through the analysis of the SVM of structural risk minimization,SVM canbuilt best consumedly,it conquer the limitation of the slumping the local trap ofPSO,and the SVM has more generalization capability than the PSO,but it has lowlearning capability than the former one.and it has tremendous influence by thepenalty factor and the kernel factor parameter,and the tradition parameterselection system has worse performing,we need exactly search a new way tooptimize it. This paper combines PSO with SVM in the foundation of analysis andcontrasting the PSO and SVM.Imposing the improving PSO to optimize thepenalty factor and kernel function,Then using the optimize SVM and model toforecast and control,and it comes into being PSO-SVM. finally,we optimize thePSO-SVM through the process of feeding supplement control model of penicillinfermentation,The result shows that the production process based on the PSO-SVM algorithm is more stable,the production is higher,and at the same time thefermentation time is shorter;Comparing the short-term power load modeling andforecasting based on PSO-SVM algorithm with conventional SVM algorithm,theresult shows that the accuracy of the load forecasting model is higher and thespeed has been improved distinctly.