【作者】 耿维；

【导师】 赵晓波； 刘建庸；

【作者基本信息】 清华大学 ， 管理科学与工程， 2010， 博士

【摘要】 本论文研究考虑行为因素的周期性盘点库存系统,包括s形效用单节点库存系统和单分销商多零售商库存系统。作为预备知识,还研究了连续需求库存系统最优策略的计算。对连续需求单节点库存系统,在离散网格的基础上建立基于数值积分和线性插值的算法以计算近似最优策略,分别构造近似最优策略成本上界和最优策略成本下界的误差估计,并证明当离散网格格点间距趋于零时近似最优策略收敛至最优策略。对S形效用单节点库存系统,分别考察指数型S形效用和一般S形效用两种情形。单峰性在对这两种情形进行理论分析时都具有重要作用。对指数型S形效用单期问题,可证明最优策略为基于初始库存水平的订达策略,其最优订达库存水平可通过搜索得到,且最优效用为单峰函数。对指数型S形效用多期问题,若需求服从对数凹分布,则可建立动态规划算法求解最优策略。对一般S形效用单期问题,若需求服从指数分布,则可证明最优策略为基于初始库存水平的订达策略,最优订达库存水平可通过搜索得到,且最优效用为单峰函数；若需求服从一般分布,则在数值算例基础上为后续研究提出若干猜想。数值算例表明,在S形效用库存系统的运作管理中,收益预期和由收益预期带来的规避风险倾向占据主导地位,而损失预期和由损失预期带来的追逐风险倾向居于次要地位。对单分销商多零售商库存系统,本论文将各种企业运作实例归纳为六种运作模式逐一建模,分别建立近似动态规划算法计算系统成本,并以数值算例比较不同运作模式下的系统绩效。计算结果表明,系统各节点只关注自身利益时的无序竞争行为严重损害库存系统整体利益,寄希望于单纯的信息共享来改善系统绩效并不现实,还需要通过合理的分配规则等多种方式来协调节点的行为。本论文所获结果有助于洞见管理者行为因素对库存系统运作管理的影响,为进一步设计应对机制以减弱甚至消除这种影响、最终达到改善库存系统运作管理绩效的目的奠定基础。更多还原

【Abstract】 This thesis investigates operations management of periodical inventory systems with considering impact of decision behaviors. Inventory system with S-shaped util-ity and single distributor multi-retailer inventory system are considered, respectively. Besides, computing optimal policy for inventory system with continuous demand is considered as a preliminary.For inventory system with continuous demand, an algorithm, in which numerical integral and linear interpolation are used, is constructed on discretized grid for com-puting near optimal policy. Upper bound of cost of the near optimal policy as well as lower bound of optimal cost are estimated then. The near optimal policy is proved to converge to the optimal policy as scale of the grid tends to zero.For inventory system with S-shaped utility, exponential-type S-shaped utility is first considered and then general S-shaped utility. Unimodality plays an important role in analysis. For a single period problem with exponential-type S-shaped utility, a state-dependent order-up-to policy with order-up-to levels found by searching is op-timal, and the optimal expected utility is unimodal. For a multi-period problem with exponential-type S-shaped utility, if demand follows log-concave distribution, a dy-namic programming algorithm can be constructed to obtain optimal policy. For a single period problem with general S-shaped utility, if demand follows exponential distribu-tion, a state-dependent order-up-to policy with order-up-to levels found by searching is optimal, and the optimal expected utility is unimodal, while if demand follows a general distribution, some conjectures on optimal policy are proposed. Numerical ex-amples reveal that expectation on gains and risk-aversion due to gains dominate the operations management of inventory system with S-shaped utility.For single distributor multi-retailer inventory system, six operations scenarios are categorized, and approximation dynamic programming algorithms for all scenar-ios each are developed to obtain system cost. Numerical examples show that irregu- lar competition among retailers due to self-interesting reduces whole system interest sharply, and that one can hardly improve system performance only by sharing informa-tion without coordinating system members by some means such as, e.g., an appropriate allocating rule.Results from this thesis provide insights on impact of managers’behaviors on operations management of inventory system, which help the managers design mech-anisms for reducing disadvantage of bounded rationality and improve performance of inventory system.更多还原