节点文献
逆向物流随机库存策略研究
Stochastic Inventory Policies for Reverse Logistics
【作者】 陈铓;
【导师】 符卓;
【作者基本信息】 中南大学 , 物流工程, 2013, 博士
【摘要】 由于回收产品在数量、时间与质量的高度不确定性,以及存在多个可替代的供应源,逆向物流的库存管理相当困难与复杂。签于传统库存优化问题的定量研究的局限性,论文采用马可夫决策过程(MDP)理论在对逆向物流库存系统的动态变化特性分析的基础上对随机库存问题进行定量研究。基于产品回收之后的形态变化,论文提出了逆向物流库存系统的两种基本类型:回收产品直接重用,回收产品再制造。在此基础上,论文将所研究的随机库存优化策略问题扩展至回收产品生命周期。首先,论文分析了回收产品直接重用的逆向物流库存系统,其中只有一个库存点,假设需求与回收之间是相互独立的,且对回收产品不进行废弃处置。针对单位需求与回收的情况,通过一维连续时间马尔可夫模型证明了库存成本函数与采用(s,Q)的传统“前向物流”库存模型的库存成本函数具有完全相同的结构。然后,针对需求与回收服从一般性随机分布的情况,分别采用贴现成本法与平均成本法证明了逆向物流库存系统的库存量可以被分解为两部分,一部分的动态特性与传统库存系统是相同的,另一部分则是独立的对外订购,此时逆向物流库存系统的最优库存策略是(s,S)库存策略。由于在一定条件下逆向物流库存模型可以转换成采用(s,S)库存策略的“前向物流”库存模型,因此论文采用了线性搜索法进行数值实验,从回收率、需求率、产品回收过程中信息控制等三方面探讨不同的回收参数对平均库存成本的影响。数值实验结果表明,如果采用合理的库存控制策略,回收率对平均库存成本的影响是相当有限的。只有在回收率非常高的时候,对回收产品进行废弃处理以避免过高的库存量才是有必要的。同时,通过预测与监测等手段加强产品回收信息的控制有助于降低逆向物流库存系统的平均库存成本。其次,论文分析了回收产品再制造的逆向物流库存系统,其中有两个库存点即回收库存、销售备用库存,假设需求与回收之间是相互独立的,且对回收产品不进行废弃处置。另外假设新产品制造与回收产品再制造共享一个服务台,再制造作业比制造作业具有优先权。论文参照Puterman提出的方法采用马尔可夫决策过程(MDP)建立了库存模型,采用单值化系数进行离散化处理,推导出最优性方程。论文首先研究了贴现成本情况下最优库存策略的结构,然后证明了平均成本情况下的最优库存策略是基于销售备用库存的最优初始库存量的库存策略。对于最优平均库存成本的求解,论文建立二维连续时间马尔可夫链截断模型,提出了基于矩阵几何法的稳态概率求解算法。为提高运算效率,论文提出了求解最优初始库存量的近似值的三种启发式算法。数值研究表明,在多数情况下,尤其是在缺货成本相对于持有成本不是很高的情况,三种启发式算法都可以得到良好地近似值。启发式2得到最优库存成本比使用完整枚举法可以减少平均约77%的计算时间。最后,论文对求解稳态概率的三种算法进行了计算时间性能的比较。数值实验表明,论文提出的矩阵几何算法是平均速度最快的。再次,在以上研究的基础上,论文探讨了逆向物流库存系统在回收产品生命周期各个阶段的最优库存策略。论文采用需求率与回收率的不同组合来代表回收产品生命周期的特定阶段,建立了马尔科夫(MDP)库存模型,对各种可能的库存策略进行数值实验以期找到最优库存策略。数值研究发现,在产品生命周期中采用何种库存策略取决于系统的持有成本组合,同时在回收产品生命周期各阶段的需求率和/或回收率发生变化的时候采用“随时变化”库存策略的设置可以有效地降低总库存成本,相对而言,如果采用“单一策略”或“部分变化”库存策略,那么将会发生额外库存成本。在回收产品生命周期各阶段的周期数量大于一个以上的时候,MDP模型所产生的库存策略可以得到近似最优库存成本。同时,随着产品生命周期各个阶段的周期数量的增加,MDP库存策略的成本性能也随之得到提高。论文提出的MDP库存策略的库存成本在大多数情况下明显低于拉式库存策略。最后,论文对逆向物流随机库存优化问题的研究成果进行了总结,并针对研究过程中的一些假设进行讨论,提出了进一步研究的扩展方向。图28幅,表23个,参考文献169篇。
【Abstract】 The highly uncertainty of return products in quantity, time and quality, as well as there are several alternative supply sources, making the reverse logistics inventory management difficult and complex. Signed in the limitations of quantitative research in traditional inventory optimization problem, this dissertation uses the Markov Decision Process (MDP) theory to analyze the dynamic nature of the reverse logistics inventory system and optimal inventory policy.Based on morphological changes in the product recovery, the dissertation proposes two kinds of the basic type of reverse logistics inventory systems:directly reuse, remanufacture. On the basis, the dissertation will extend stochastic inventory optimization to the return product life cycle.First, the dissertation studies the reverse logistics inventory system with return product directly reuse, in which only one inventory point, assuming that demand and return are independent of each other, without disposal of return product. For the unit demand and return, the dissertation construts one-dimensional continuous-time Markov model to prove that the inventory cost function has exactly the same structure as inventory cost function of the traditional inventory model with (s, Q) inventory policy. Then, subject to the general random distribution for demand and return, the dissertation proves the inventory level of the reverse logistics inventory system can be broken down into two parts, the dynamic characteristics of part is the same as the traditional inventory system, the other part is independently the external order, so the optimal inventory policy of reverse logistics inventory system is (s, S) inventory policy.Because of reverse logistics inventory models can be converted into forward logistics inventory model with (s, S) inventory policy under certain conditions, some traditional inventory optimization algorithms can also be used in the reverse logistics inventory system. The dissertation investigates different recovery parameters affecting the average inventory cost by linear search method proposed by Zheng&Federgruen three areas of demand rate, return rate and product retrun process. The numerical results show that, if we adopt a reasonable inventory control policy, return rate is fairly limited impact on the average inventory cost. Only when return rate is very high, it is necessity to disposal return product in order to avoid excessive inventory levels. At the same time, the dissertation pointed out to control the product return information by means of prediction and monitoring that can reduce the average inventory cost of reverse logistics inventory system.Second, the dissertation studies the reverse logistics inventory system with return product remanufacture, which has two inventory point of recovery inventory and serviceable inventory. The dissertation assumes that demand and return are independent of each other, without disposal of return product. The dissertation consideres a hybrid production and remanufacturing system where both production and remanufacturing operations are performed by the same single server, and remanufacturing operation has priority. This dissertation construtes the inventory model by Markov Decision Process (MDP), which uses the coefficient of uniformization to discretization, so the optimality equation is derived. This dissertation firstly studies the structure of optimal inventory policy in the discount cost situatio, then proves that the optimal inventory policy is the inventory policy that based on the optimal initial inventory level of serviceable inventory in the average cost situatio.For the optimal average inventory cost, the dissertation constructs two-dimensional continuous-time Markov chain truncated model, and proposes a solution algorithm based on matrix-geometric methods to solve steady-state probability. In order to improve computational efficiency, the dissertation proposes three heuristic algorithms for solving the approximation of the optimal initial inventory level, in which heuristic1,2are based on the precise formula of the production queuing system with product return, heuristic3is a combination of both. Numerical studies have shown that in most cases, especially in the shortage cost is not very high, all three heuristics can obtain the good approximate value. Heuristics2may reduce approximately77%computing time to obtain the optimal inventory cost than the complete enumeration method. Finally, the dissertation compares the computing time performance of three algorithms for solving steady-state probability. Numerical experiments show that the matrix geometric algorithm proposed by the dissertation has the fastest average speed.Once again, on the basis of the above studies, the dissertation discusses optimal inventory policy of reverse logistics inventory system in the various stages of the return product life cycle. The dissertation uses the demand rate and the returns-ratio different combination to represent the specific stage of the return product life cycle, establishes a Markov inventory model, carries on the value experiment to find the optimal inventory policy on a variety of possible inventory policies. The results indicate that the optimal inventory policy structures to be used over the return product life cycle depend on the holding cost structure of the system. The optimal parameter values of the inventory policies are sensitive to the changes in demand rate and return rate, and the cost of an inventory policy is sensitive to the changes in its parameters values. Hence, it is recommended to revise the inventory policies over the return product life cycle every time a change in demand rate and/or return rates occurs. Instead of frequently revising the inventory policies over the life cycle, if no revision or only a partial revision is done, then a significant amount of additional cost occurs. The optimal inventory policies based on the MDP model are also good approximations of the optimal policies in a finite-horizon life cycle setting if the life cycle stages are at least a few periods long. Clearly, as the number of periods in which a certain pair of demand rate and return rate is valid increases, the performance of the optimal inventory policy improves.Finally, the dissertation summarizes the study results of the stochastic inventory optimization problem in reverse logistics, and discusses some assumptions in the course of the study, proposes the extension directiones for further research. The dissertation contains28figures,23table and169references.
【Key words】 Reverse Logistics; Inventory Policy; Stochastic; MarkovDecision Processes; Steady State Probability;