论文部分内容阅读
传统固定尺度的核函数模型不适合稀疏地表示震荡信号.为了提高震荡信号表示的稀疏性,提出了一种尺度可调的核函数模型的建立方法.该方法通过正交最小二乘算法进行逐步回归建模,选择每一个回归子时,利用群搜索算法优化残差目标函数,计算相应的核函数的尺度.实验结果表明,可调核函数模型比传统的固定尺度核函数模型具有更强的稀疏性和泛化能力.
The traditional fixed-scale kernel function model is not suitable for the sparse representation of the oscillatory signal.In order to improve the sparsity of the oscillatory signal representation, a scalar-adjustable kernel function model is proposed, which is solved by orthogonal least square algorithm Regression modeling, the choice of each regression sub-group, the use of group search algorithm to optimize the residual objective function to calculate the scale of the corresponding kernel function.Experimental results show that the adjustable kernel function model than the traditional fixed-scale kernel function model has a stronger Sparsity and generalization ability.