基于PSO算法的SVM参数优化方法研究

Method of Parameters Optimization in SVM based on PSO

为了克服支持向量机中核函数参数的不确定性及解决核参数的最优选择问题，本文将起源于人工生命和演化计算理论的粒子群优化算法运用到支持向量机的参数选择中。在对基本粒子群优化算法工作原理分析的基础上，对基本算法的收敛速度进行了适度改进，使其具有自适应能力，即在初期进行快速搜索，而在末期进行精细搜索，从而扩大参数搜索的宽度和深度，满足多样化和集中化的特点。仿真实验表明，通过该方法选择出来的核参数能够提高分类及预测精度，具有实用性。

To overcome the uncertainty and to resolve the problem of parameters optimization in kernel function of support vector machine （SVM）, particle swarm optimization （PSO） method, which was originated form artificial life and evolutionary computation, is applied to SVM＇s parameters selection and optimization in the paper. The improved PSO algorithm of increasing convergence rate is proposed based on the analyzingprinciple of basic PSO. Thereupon, the improved PSO algorithm has self-adaptive ability that can be faster searching in early phase and more carefully searching in latter phase rather than basic PSO, and can be meeting the requests of diversification and intensification. The simulation experiment results demonstrate that, the selected kernel parameters by the new PSO algorithm can improve the overall performance of the SVM classifier and have new application domain.