基于捕食搜索策略遗传算法的SVM参数优化方法

摘要

基于支持向量机(SVM)模型的泛化能力和拟合精度与其相关参数的选取有关,提出将捕食搜索策略的遗传算法(PSGA)运用到SVM的参数选取中.该算法以最小化输出量的拟合误差为目标,以SVM的3个参数作为决策变量.通过对谷氨酸发酵过程建模的实验表明,该方法可以提高谷氨酸浓度的训练精度及预测精度,是一种优化SVM参数的有效方法.%Based on the fact that generalization and fitting accuracy of the Support Vector Machine (SVM) model depend on its parameters setting, a predatory search genetic algorithm was proposed to determine the parameters of the SVM. The target of this algorithm was to minimize the fitting error of output and three parameters of SVM were used as the decision variables. An application example on the glutamic acid fermentation process shows that the method can improve the training accuracy and forecast accuracy of glutamate concentration and is an effective way to optimize the parameters of SVM.