基于Cholesky分解的LSSVM在线学习算法  

An online learning algorithm for LSSVM based on Cholesky factorization

作　　者：蒋星军[1,2] 周欣然[3] 唐钊轶[2] 

机构地区：[1]北京工业大学计算机学院,北京100022 [2]湖南广播电视大学计算机系,湖南长沙410004 [3]中南大学信息科学与工程学院,湖南长沙410075

出　　处：《湖南科技大学学报（自然科学版）》2013年第4期74-77,共4页Journal of Hunan University of Science And Technology：Natural Science Edition

基　　金：国家自然科学基金(10971060);湖南省科学技术厅项目(2011FJ6033)

摘　　要：针对最小二乘支持向量机(LSSVM)用于在线建模时存在的计算复杂性问题,提出一种LSSVM在线学习算法.首先引入了基于Cholesky分解求LSSVM的方法,接着根据在线建模期间核函数矩阵的更新特点,将分块矩阵Cholesky分解用于LSSVM的在线求解,使三角因子矩阵在线更新从而得出一种新的LSSVM在线学习算法.该算法能充分利用历史训练结果,减少计算量.仿真实验显示了这种在线学习算法的有效性.Aiming at the computational complexity of least squares support vector machine （LSSVM） ＇ s online modeling, an online learning algorithm for LSSVM was proposed. First, the solution of LSSVM through the Cholesky factorization was introduced, then the Cholesky factorization of partitioned matrix was applied to the online solution of LSSVM according to the updating character of kernel function matrix during online modelling, and triangle factor matrix was renewed online, consequently, a novel online ]earning algorithm for LSSVM was obtained. The improved learning algorithm can make full use of the historical training results and reduce the computation amount. The numerical simulation results the validity of the online learning algorithm for LSSVM.

关 键 词：最小二乘支持向量机 在线学习 CHOLESKY分解 滚动时间窗 系统在线辨识 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP273[自动化与计算机技术—控制科学与工程]