一种基于Cholesky分解的动态无偏LS-SVM学习算法

Dynamic non-bias LS-SVM learning algorithm based on Cholesky factorization

针对最小二乘支持向量机用于在线建模时存在的计算复杂性问题,提出一种动态无偏最小二乘支持向量回归模型.该模型通过改进标准最小二乘支持向量机结构风险的形式消除了偏置项,得到了无偏的最小二乘支持向量机,简化了回归系数的求解.根据模型动态变化过程中核函数矩阵的特点,设计了基于Cholesky分解的在线学习算法.该算法能充分利用历史训练结果,减少计算复杂性.仿真实验表明了所提出模型的有效性.

Aiming at the computational complexity of LS-SVM＇ s on-line modeling, a dynamic non-bias least square support vector regression model is proposed. The model eliminates the bias of LS-SVM by improving the form of structure risk. As a result, the non-bias LS-SVM is achieved and the calculation method of regression coefficients is simplified. Then an online learning algorithm based on the Cholesky factorization is designed according to the character of kernel function matrix in the model＇s dynamic change process. The improved learning algorithm can make full use of the historical training results and reduce the computational complexity. Experimental results indicate the effectiveness of the dynamic non-bias LS-SVM.