混沌时间序列的最小二乘支持向量机预测

提出了最小二乘支持向量机混沌时间序列预测方法,并研究了三种混沌信号的预测性能。该方法在优化指标中采用了平方项,且只有等式约束,将传统支持向量机求解二次规划问题转化为求解线性方程组,因而简化了计算复杂性。仿真实验结果表明该方法预测模型参数选择容易、在较大范围内取值时对预测误差影响很小,而且即使在输入维数m小于Takens嵌入定理所确定的维数时,也具有很好的预测性能。

Prediction of chaotic time series using least square support vector machines

A novel least square support vector machines predictive approach is presented and applied to complete the research of the predictive performances of three chaotic time series.This approach exploits the quadratic term in optimization index mark and only contains the equality constraint.It transforms the resolution of the quadratic programming of the classical support vector machines to the resolution of the linear system of equations.As a result,the computational complexity is simplified.The experimental results indicate that the parameters for the predictive model are selected easily and impact the predictive error slightly in larger range.Even though the input dimension m is smaller than the dimension determined by Takens embedded theorem,the proposed approach has good predictive performances.