Research On Support Vector Machine Training Algorithm Based On Simplex Evolution

Support Vector Machines(SVM)is a new learning method based on statistical learning theory.However,the generalization ability of one class of support vector machines depends largely on the selection of kernel functions.According to the current research status of support vector machine,it is found that the introduction of intelligent optimization algorithm to optimize the parameters of support vector machine has certain limitations.For example,some intelligent optimization algorithms rely too much on the control of parameters in optimization,and the change of control parameters has a great impact on the classification and recognition of support vector machines.Firstly,this paper studies support vector machine,and finds that the advantages of neural network can be better applied to support vector machine and improve its generalization ability.Therefore,this paper first studies the support vector machine based on neural network.With the deepening of research,it is found that support vector machine based on neural network is more or less unstable in parameter optimization.Because of the large number of control parameters,it is difficult to control and modify,and the result can not reach the expected value.For this reason,this paper quotes the Surface-Simplex Swarm Evolution(SSSE),which is a new swarm intelligence optimization algorithm.The single control parameter characteristic of the simplex evolutionary optimization algorithm improves the reliability and universality of the algorithm,balances the pertinence and diversity of the group search through the multi-role characteristics of the algorithm,improves the search efficiency and convergence performance of the training algorithm,guarantees the generality of the algorithm,and verifies the effectiveness of the training algorithm.Finally,the training algorithm based on Surface-Simplex Swarm Evolution support vector machine is tested and validated.Through the application of data,the weights,thresholds and kernel functions of the neural network in the new support vector machine are optimized,and the global optimal solution is obtained,which improves the recognition rate.Three groups of database data were used to simulate the data by selecting Gauss kernel function and polynomial kernel function respectively,and the experimental results were analyzed.The simulation results show that compared with other algorithms,the SVM trained by this algorithm not only improves the recognition rate effectively,but also reduces the influence of control parameters on learning performance,and improves the universality and robustness of the algorithm in application.The application of support vector machines based on Surface-Simplex Swarm Evolution in reality has a certain development.