再生核支持向量机在非线性系统中的应用

Reproducing Kernel Function Based on Walsh Series

为了提高非线性系统辨识的精度,提出用Walsh函数作为空间V0的尺度函数,构造出L2(R)空间的正交规范序列。结合小波多分辨分析,将Hilbert空间分为一系列子空间,并由可分Hilbert空间与L2(R)的等价性,利用内积同构的线性算子,可以把V0子空间的尺度函数折算为Hilbert空间的子空间V0的尺度函数,构造出新的Walsh序列再生核。通过仿真实验,与传统的RBF核函数、高斯核函数等比较,该尺度再生核函数具有更高的辨识精度,较少支持向量数目,充分体现了支持向量机较好的推广性能。

A new reproducing kernel function of least square support vector machines based on Walsh series is presented in this paper. The reproducing kernel is constructed in reproducing kernel Hilbert space （RKHS）, Because the Hilbert space and the square integrable space are isomorphic, according to the wavelet multi-resolution analysis, Walsh consequence can be seen a set of orthogonal basis to construct the reproducing kernel. The simulation results are discussed to illustrate the proposed method.