基于支撑向量回归(SVR)可以通过构建支撑向量机分类问题实现的基本思想,推广最小类方差支撑向量机(MCVSVMs)于回归估计,提出了最小方差支撑向量回归(MVSVR)算法.该方法继承了MCVSVMs鲁棒性和泛化能力强的优点,分析了MVSVR和标准SVR之间的关系,讨论了在散度矩阵奇异情况下该方法的求解问题,同时也讨论了MVSVR的非线性情况.实验表明,该方法是可行的,且表现出了更强的泛化能力. Based on the support vector regression (SVR), the basic idea of ​​constructing support vector machine (SVM) classification problem can be constructed and the least square support vector regression (MVSVR) algorithm is proposed by using least squares support vector machines (MCVSVMs) MCVSVMs has the advantages of robustness and generalization ability, analyzes the relationship between MVSVR and standard SVR, discusses the solution of this method in the case of divergence matrix singularity, and also discusses the MVSVR nonlinearity.Experiments show that , This method is feasible and shows a stronger generalization ability.