新的基于预估校正法的大规模营销优化算法  

New large-scale marketing optimization algorithm based on predict-correct

作　　者：吴敏[1] 王喜玲[1] 何蓓[1] 

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《系统工程理论与实践》2009年第9期160-172,共13页Systems Engineering-Theory & Practice

基　　金：国家杰出青年科学基金(60425310);国家863计划(2006AA04Z172)

摘　　要：针对大规模一对一营销问题,提出一种时间复杂度为O(nlogn/ε)(n为决策变量数,ε为允许误差)的大规模一对一营销优化算法.它基于预估校正思想,在预估、校正步长计算中采用LDL分解,并结合列近似最小度排序算法,有效降低时间复杂度.同时,算法在预估步中引入步长参数,根据步长参数值自适应更新中心参数,使得算法具有超线性收敛性.实际测试表明,该算法可在短时间内精确求解10万以上客户规模的一对一营销优化问题.Aimed at a large-scale one-to-one marketing optimization problem, it is presented a largescale one-to-one marketing optimization algorithm with the time complexity 0（n log n/ε）（n is the number of decision-making variables and ε is the permissible error）. It bases on predict-correct linear program method, uses LDL faztorization in predict-correct calculation, and uses column approximate minimum degree ordering algorithm to effectively reduce time complexity. Meanwhile, this algorithm uses step parameter in predict step, and the value of the step parameter to calculate center parameter adaptively, which makes sure that it has superlinear convergence. It is shown in the test that this algorithm can solve the one-to-one marketing optimization problem accurately in a short time, which the number of customers is 100,000 or above.

关 键 词：一对一营销优化 预估校正法 列近似最小度排序 LDL分解 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]