尺度核支持向量机及在动态系统辨识中的应用

为了提高非线性动态系统辨识质量,提出了新的支持向量机尺度核函数构造方法.首先直接构造紧支撑尺度函数,然后根据小波多分辨分析理论,由紧支撑尺度函数生成具有多分辨率特性的尺度核函数.证明了这种核函数是满足Mercer定理的支持向量机核函数.动态系统辨识的仿真结果表明,尺度核函数支持向量机的建模和逼近能力优于基于三阶样条核函数或RBF核函数的支持向量机.

Support Vector Machine with Scaling Kernel and Its Application in Dynamic System Identification

To improve the performances of dynamic system identification,a new method to formulate scaling kernel function for support vector machine was proposed.In the method,a compact support scaling function is derived first,and then based on wavelet multiresolution analysis a scaling kernel function with multiresolution characteristics is constructed from the compact support scaling function.It was proved that this scaling kernel function satisfies Mercer conditions and can be used as a kernel function for support vector machine.Simulation results show that the support vector machine with the proposed scaling kernel function has better modeling and approximation abilities than that with a Spline function kernel or RBF(radial basis function) kernel.