一种大样本学习最小二乘支持向量回归模型  

A Least Squares Support Vector Regression Model for Learning of Large Samples

作　　者：袁从贵[1] 徐淑琼[1] 张新政[2] 

机构地区：[1]东莞职业技术学院，广东 东莞 523808 [2]广东工业大学自动化学院，广州 511442

出　　处：《控制工程》2017年第9期1768-1773,共6页Control Engineering of China

基　　金：国家自然科学基金(61074185);广东省自然科学基金(2015A030310257;2014A030310380)

摘　　要：针对最小二乘支持向量回归大样本学习效率偏低的问题,提出了一种最小二乘支持向量回归快速学习算法模型。首先将欧氏距离进行推广,设计了一种支持向量回归高维特征空间相似性测度标准,然后构建了无监督核聚类分析支持向量选择算法,再通过Nystr?m方法逼近原最小二乘支持向量回归学习问题的解。最后采用Sinc函数和多个数据集测试了模型的性能。实验结果表明,在预测误差没有明显下降的情况下,该模型能克服最小二乘支持向量回归处理大样本学习问题时的内存溢出错误,显著提高其学习效率。In order to improve the efficiency of least squares support vector regression （LS-SVR） for learning of large samples, a fast LS-SVR model is put forward here. Firstly, the Euclidean distance is extended, and designed as a similarity metric in the high dimensional feature space of support vector regression. Then, the unsupervised kernel clustering algorithm is developed to select support vectors. Furthermore, the Nystrrm method is performed to approximate the solution of the LS-SVR. Finally, the performance is tested by Sinc function and several examples. The experimental results demonstrate that the proposed model can overcome the memory overflow error, and improve the learning efficiency significantly with desired accuracy.

关 键 词：大样本学习 最小二乘支持向量回归 核聚类 Nystr?m逼近 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]