基于差分进化算法的支持向量回归机参数优化

摘要

支持向量机是结构风险最小化原理的一种新型学习技术,被广泛应用到很多工业控制领域中,良好的泛化能力和预测精度在很大程度上受到参数选取的影响.传统参数选择方法易陷入局部最优,为提高优化识别参数的精度和效率,提出基于差分进化算法的支持向量回归机参数优化算法.以均方误差最小为优化准则,差分进化算法的全局寻优能力,搜索支持向量回归机的最优参数组合,达到对参数的最优选择.通过Matlab进行仿真实验,结果表明改进的算法不仅加快参数搜索和优化的速度,而且选择的最优参数能大大提高支持向量机预测精度和泛化能力,并具有良好的鲁棒性和较强的全局寻优能力.%Support vector machine is a learning technology based on structure risk minimization and has been widely used in many industry applications, the good generalization ability and estimation accuracy are impacted by parameters selection of SVM. Traditional parameter optimization methods are easy to fall into local optimum and of low optimizing efficiency. Parameters optimization of SVM based on differential evolution is presented to solve these problems. Optimal rule is the least mean square error of samples, and the combination of SVM parameters are optimized based on differential evolution algorithm for global optimization ability. Algorithm results are compared with the other algorithms with Matlab. The simulation results show that the algorithm can accelerate the speed of parameters searching and improve the prediction accuracy of SVR. It is of good robustness and strong global search capability.