基于SVM的大样本集系统辨识与函数拟合仿真  

The Simulation of Large Scale Regression and System Identification Problems Based on Support Vector Machine

作　　者：张智[1] 朱齐丹[1] 邢卓异[1] 

机构地区：[1]哈尔滨工程大学自动化学院,黑龙江哈尔滨150001

出　　处：《计算机仿真》2006年第9期123-128,共6页Computer Simulation

摘　　要：支持向量机是一种优秀的学习方法,也是具有很好泛化性能的回归方法。但由于支持向量机算法实习复杂,效率低,严格限制了其应用,SMO算法的提出大大提高了支持向量机的学习效率。因此,借助SMO算法,便可以实现大样本集的非线性系统辨识和函数拟合。文中对回归问题的SMO算法作了详细介绍,并对其进行改进。然后研究了利用改进SMO算法的非线性系统辨识方法,给出了非线性系统的辨识的仿真结果,和一维二维函数的拟合仿真。并通过仿真,与原始算法进行了比较,显示了改进SMO算法的快速性。Support Vector Machine is an excellent learning technique, and it is also a class of regression method with a good generalization ability. However, because the algorithm of SVM is complex, it hinders the application of SVM, the SMO algorithm improves the learing velocity of SVM greatly. So, using SMO learning method, we can solve the problems of large scale system identification and regression. In this paper, the regression SVM algorithm is introduced, and the method is improved from many respects. The method of nonlinear system identification is studied. The simulation results of the improved SMO method and the origin SMO method show the advantages of the improved SMO method.

关 键 词：支持向量机 序列最小优化 改进学习算法 回归问题 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]