【作者】 施文；

【导师】 刘志学；

【作者基本信息】 华中科技大学 ， 管理科学与工程， 2013， 博士

【摘要】 在实地调研、总结和评述相关文献的基础上,本论文深入分析了在国内众多汽车制造商产能扩大背景下的第三方物流企业(3PL)运作决策优化问题。首先建立符合零部件供应物流实际运作的离散(随机动态)仿真模型,即基于3PL的零部件循环取货越库配送(3PL-MRCD)模型；然后,探讨了产能扩大对3PL-MRCD物流绩效(零部件平均物流时间及物流量)的重要影响；最后,深入研究了3PL-MRCD系统的仿真优化设计及稳健(仿真)优化设计。具体而言,论文主要完成的创新性工作总结如下：建立零部件从始端循环取货、中端越库配送直到终端线边仓库的供应物流3PL-MRCD仿真模型,通过灵敏度分析探讨了汽车制造商产能扩大对系统物流绩效的影响,在此基础上提出了四种可供3PL选择的物流绩效改进方案,包括车辆调度方法、车辆排队规则、越库中心库台分配模式及不同车辆类型。提出针对复杂系统的多重响应序贯分支因子筛选设计方法(MSB)。鉴于3PL-MRCD系统的复杂性(存在较多的因子),提出首先筛选出影响系统的关键因子、再针对关键因子进行优化的研究步骤。由于所研究的响应(系统绩效或仿真输出)具有多维性,提出一个针对多重响应的序贯分支筛选法,MSB将传统序贯分支法(SB)拓展到多重响应领域,蒙特卡洛实验表明MSB比SB更具效率和效力,并最终应用到3PL-MRCD系统中,筛选出影响系统绩效最重要的少量因子。基于响应面(RSM)和Kriging元模型两类仿真优化方法对筛选出的关键因子进行优化(MSB-RSM和MSB-Kriging),确定其最大化系统绩效的最优参数水平。在优化结果的基础上,对比分析了产能扩大后3PL可能的四种物流决策方案,确定了物流绩效改进的最佳思路,并比较分析了MSB-RSM和MSB-Kriging两类方法拟合的优劣以及传统RSM和MSB-RSM方法的效率与效力。采用拉丁超立方采样(LHS)与RSM和Kriging相结合的仿真优化方法对3PL-MRCD系统进行稳健优化设计。利用LHS在环境因子若干等概率空间中随机样本点与筛选出的关键因子进行交叉设计,最终确定在最小化系统绩效变动的参数水平。在优化结果的基础上,应用自由分布的Bootstrap法对比分析了仿真优化与稳健优化方法最优解的优劣。更多还原

【Abstract】 On the basis of field observations and literature review, this dissertation deeply examines the best operating settings of Chinese third-party logistics under the expansion of their serving manufacturer. In the proposed framework, a random (stochastic) dynamic model that agrees with the3PL’s practical operations more; i.e., milk-run pickup and cross-docking distribution (3PL-MRCD for short) is constructed, and then changes of the3PL-MRCD performances (cycle time, CT and number of throughput, NT) due to the addition of the assembly plant capacity are explored. Two types of metamodel-based optimization routines simultaneously---simulation optimization methods and robust (simulation) methods---to identify factor levels that would maximize system potential are employed. More specifically, the main contributions of the dissertation are as follows.To understand the changes of the3PL-MRCD system after the expansion auto maker, a process-centric modeling paradigm that ranges from the start milk-run pickup, middle cross-docking distribution, to the end factory warehouse is contructed using Arena software. Next, by conducting sensitivity analysis, the changes of system performances are observed, and logistics modes; namely, truck scheduling, truck dispatching rule, door assignment and truck type selection to improve system performance of interest are provided.To optimize the3PL-MRCD system, which involves a large number affecting factors, a hybnrid optimization framework that first searches for a few important factors dominating the system performance, and next identify the best settings of them is innovatively presented. Since the3PL-MRCD system contains two types of response; i.e., CT and NT, a novel method for factor screening in random discrete-event simulation with multiple response types, called multiple sequential bifurcation (MSB) is introduced. This MSB extends basic sequential bifurcation (SB) to incorporate multiple responses. The performance of MSB is proven and compared with the original SB procedure in two Monte Carlo experiments. Finally, the MSB is successful in eliminating the unimportant ones among numerous factors within the3PL-MRCD system, which further exhibits its robustness.To determine the optimal levels of these key factors found by MSB, two types of metamodel-based approaches; namely, response surface methodlogy (RSM) and Kriging are adopt. These metamodels are capable of describing the relationship between key factors and multiple system responses of interest, and ultimately enable optimization. After that, simulation outputs from four input combinations, which represent four decisions the3PL may make are compared objectively, and the most desired way to improve system performance after the expansion of the assembly plants’capacity is then provided. In addition, the fitness between MSB-RSM and MSB-Kriging methods, and also the efficiency and efficacy between classical RSM and our MSB-RSM are compared.To discern the robust conditions for the key factors found by MSB, a robust procedure that combines the Latin hypercube sampling (LHS) with RSM and Kriging methods is presented, and an environmental factor is taken into account. LHS can account for the distribution of the environmental factor, and form a cross (combined) design with the decision (key) factors. After that, this design to build LHS-RSM and LHS-Kriging metamodels is exploited, respectively, and find the robust operation conditions for decision factors that minimize the variability of two responses while ensure their mean values stay with a desired targets. Finally, a distribution-free bootstrapping procedure to further compare the results obtained by the simulation optimization and the robust optimization methods are presented.更多还原