摘要
比较了现今应用比较广泛的3种支持向量机(SVM)参数优化方法.具体分析了网格法、遗传算法和粒子群算法在SVM参数优化方面的性能以及优缺点,提出了一种改进的网格法.先在较大范围内进行搜索,在得到的优化结果附近区域再进行精确搜索.实验表明改进的网格搜索法耗时短,更适用于有时间要求的说话人识别应用中.
Three kinds of SVM parameters optimization methods were compared in this paper. The performances and characteristics of grid search, genetic algorithm and particle swarm optimization in SVM parameters optimization were analyzed. In this paper an improved grid search method was proposed. Firstly, we searched a set of parameters in a large space and then searched accurately around the parameters we had found. The simulation shows that the improved method ueses less time and is suitable for the applications of speaker recognition limited by time.
出处
《应用科技》
CAS
2012年第3期28-31,共4页
Applied Science and Technology
关键词
支持向量机
参数优化
网格搜索
遗传算法
粒子群算法
说话人识别
support vector machines
parameters optimization
grid search
genetic algorithm
particle swarm optimization
speaker recognition