计算机工程与设计
計算機工程與設計
계산궤공정여설계
COMPUTER ENGINEERING AND DESIGN
2014年
7期
2541-2544,2611
,共5页
最小二乘支持向量机%原空间%线性系统%三角分解%共轭梯度法
最小二乘支持嚮量機%原空間%線性繫統%三角分解%共軛梯度法
최소이승지지향량궤%원공간%선성계통%삼각분해%공액제도법
LS-SVM%primal space%linear system%triangular factorization%conj ugate graduate
最小二乘支持向量机在对偶空间训练,而原空间优化的近似解优于对偶空间优化的近似解,为此构造原空间最小二乘支持向量机(primal least square support vector machine,PLS-SVM)。将等式约束纳入目标函数构造无约束优化模型,根据最优解条件导出线性系统。对核矩阵进行三角分解,将线性和非线性最小二乘支持向量机的训练归结为相同形式,利用共轭梯度法求解。对比SVM和LS-SVM的仿真结果表明,PLS-SVM具有最高的分类精度和最短的训练时间。
最小二乘支持嚮量機在對偶空間訓練,而原空間優化的近似解優于對偶空間優化的近似解,為此構造原空間最小二乘支持嚮量機(primal least square support vector machine,PLS-SVM)。將等式約束納入目標函數構造無約束優化模型,根據最優解條件導齣線性繫統。對覈矩陣進行三角分解,將線性和非線性最小二乘支持嚮量機的訓練歸結為相同形式,利用共軛梯度法求解。對比SVM和LS-SVM的倣真結果錶明,PLS-SVM具有最高的分類精度和最短的訓練時間。
최소이승지지향량궤재대우공간훈련,이원공간우화적근사해우우대우공간우화적근사해,위차구조원공간최소이승지지향량궤(primal least square support vector machine,PLS-SVM)。장등식약속납입목표함수구조무약속우화모형,근거최우해조건도출선성계통。대핵구진진행삼각분해,장선성화비선성최소이승지지향량궤적훈련귀결위상동형식,이용공액제도법구해。대비SVM화LS-SVM적방진결과표명,PLS-SVM구유최고적분류정도화최단적훈련시간。
The proximal optimal solution in the primal space is better than that in the dual space,while least square support vector machine trains in the dual space.An algorithm PLS-SVM was proposed in the primal space.Equality constraints were integrated into the objective function to construct an unconstrained optimization model,and the optimization condition was uti-lized to derive the linear system.Triangular factorization was run for the kernel matrix and the linear and nonlinear LS-SVM training were unified as the same formulation,and the conj ugate graduate method was utilized to figurate out the optimal solu-tion.Comparisons with SVM and LS-SVM showed that,PLS-SVM had the highest classification accuracy and the lowest trai-ning time.