Block Downdating of Least Squares Solutions
Block Downdating of Least Squares Solutions
复制标题
DOI：
10.1137/s089547989223691x
复制
发表时间：
1994-07
期刊：
影响因子：
--
通讯作者：
L. Eldén;Haesun Park
中科院分区：
文献类型：
--
作者：
L. Eldén;Haesun Park
本文介绍了一种新的算法，该算法将LINPACK降级算法用于单行降级，以一种有效的方式对行块降级。然后将修正后的半正则方程的方法应用于LINPACK型块降维算法，以产生精确的降维解。给出了Cholesky块更新问题的灵敏度分析。在此基础上，提出了一种混合算法，该算法具有LINPACK型算法计算量小和修正的半正则方程(CSNE)块更新算法精度高的优点。给出了三种新的递推最小二乘滑动窗法块降维算法的精度比较的数值试验结果。
This paper introduces new algorithms that extend the LINPACK downdating algorithm for a single row downdating, to downdating of a block of rows in an efficient way. The method of the corrected seminormal equations is then applied to the LINPACK-type block downdating algorithm to produce accurate downdated solutions. A sensitivity analysis of the Cholesky block downdating problem is presented. Based on this analysis, a hybrid algorithm is developed that has the advantages of the lower computational cost of the LINPACK-type algorithm and the higher accuracy of the corrected seminormal equation (CSNE) block downdating algorithm. Numerical test results comparing the accuracy of these three new block downdating algorithms for the recursive least squares sliding window method are presented.