基于柯西分布加权的最小二乘支持向量机  

Weighted least square support vector machine based on Cauchy distribution

作　　者：邢永忠[1,2] 吴晓蓓[1] 徐志良[1] 

机构地区：[1]南京理工大学自动化学院,南京210094 [2]中国人民解放军63853部队,吉林白城137001

出　　处：《控制与决策》2009年第6期937-940,共4页Control and Decision

摘　　要：针对Suykens等提出的加权最小二乘支持向量机(WLS-SVM)回归建模的不足和防止辨识模型的"过拟合",利用柯西分布函数的一些特性,提出了基于柯西分布加权的最小二乘支持向量机.根据预测误差的统计特性,以确定加权规则的参数,从而赋予训练样本不同的权值.由于考虑了生产过程中样本的实际特性,与已有的加权方法相比,新的加权最小二乘支持向量机更具有鲁棒性.仿真结果验证了该方法的可行性和有效性.Suykens et al. （2002） presented a weighted least-squares support vector machine （WLS-SVM） for regression problems and a weighted algorithm for robust approximation under the model bestirred enough. Therefore, by using the characters of Cauchy distribution function, a least-squares support vector machine based on Cauchy distritution weight is presented, which is prone to over-fitting under the model bestirred deficiently. Different values for weighted factor are selected based on statistical features of the prediction error. For the real features of the samples in the proceeding of production, the new WLS-SVM is more robust. The result of a numerical regression experiment shows the feasibility and effectiveness of this algorithm.

关 键 词：加权最小二乘支持向量机 柯西分布函数 过拟合 预测误差 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]