Recursive regularisation parameter selection for sparse RLS algorithm
Recursive regularisation parameter selection for sparse RLS algorithm
复制标题
稀疏RLS算法的递归正则化参数选择
DOI：
10.1049/el.2017.4242
复制
发表时间：
2018
影响因子：
1.1
通讯作者：
Zhang Youwen
中科院分区：
文献类型：
--
作者：
Sun Dajun;Liu Lu;Zhang Youwen
在这封信中，作者提出了一个递归正则化参数选择方法稀疏递归最小二乘（RLS）算法。建议的RLS算法是正规化的凸函数，这等于两个凸函数的线性组合，一个是为了科普随机稀疏，另一个是为了科普组稀疏。推导了相应于RLS算法的正规方程和凸正则化罚函数，并提出了一种更新正则化参数（即线性组合的系数）的递归算法。作为一个例子，通过使用l 0 -范数和l 2，0 -范数的线性组合作为惩罚函数，仿真结果表明，对于同时具有随机稀疏性和群稀疏性的慢时变稀疏系统，所提出的具有递归正则化参数选择的稀疏RLS在均方误差方面具有更好的性能。
In this Letter, the authors propose a recursive regularisation parameter selection method for sparse recursive least squares (RLS) algorithm. The proposed RLS algorithm is regularised by a convex function, which equals the linear combination of two convex functions, one to cope with random sparsity, and the other to cope with group sparsity. The normal equations corresponding to the RLS algorithm with the proposed convex regularised penalty function are derived, and a recursive algorithm to update the regularisation parameters (i.e. the coefficients of the linear combination) is proposed. As an example, by using the linear combination of an l 0 -norm and an l 2 , 0 -norm as the penalty function, simulation results show that the proposed sparse RLS with recursive regularisation parameter selection can achieve better performance in terms of mean square error for a slowly time-varying sparse system with both random sparsity and group sparsity.