一种基于消去树的LDL分解方法及其在营销优化计算中的应用  

Elimination-Tree-Based LDL-Factorization and its Application to Marketing Optimization Computation

作　　者：何蓓[1] 吴敏[1] 桂卫华[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《小型微型计算机系统》2006年第10期1975-1979,共5页Journal of Chinese Computer Systems

基　　金：国家杰出青年科学基金项目(60425310)资助;教育部青年教师奖项目(教人[2002]5号)资助

摘　　要：基于分解因子消去树理论,提出一种新的LDL分解计算方法,它利用消去树结构,将LDL分解过程分为非零元素结构预测和数值计算两部分,从而有效避免多余的内存开销和数值运算.并给出其应用与OnetoOne营销优化算法的实现方法.测试结果表明,该方法能以较少的乘法运算次数快速确定分解因子,从而大幅度提高OnetoOne营销优化算法的性能.This paper presents a new implementation method for LDL factorization based on the theory of elimination tree for factorization factor. With elimination tree theory, this method breaks the computation of LDL factorization into two stages, predicting the nonzero structure of the factor and computing the numerical value of the nonzero element, so that it can avoids unnecessary memory cost and numerical operation. And apply it in One to One marketing algorithm. The experimental results on practical problems indicate that this method can compute the LDL factor by the lesser number of multiplication computa- tion, so it can obviously improve the performance of solving the problem of One to One marketing optimization.

关 键 词：ONE to One营销优化 稀疏矩阵 消去树 LDL分解 结构预测 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]