[转帖]2006全国优秀博士论文道路交通领域唯一一篇

itslu

文道路交通领域唯一一篇，也是目前道路交通领域优秀博士论文第2篇
作者姓名：孙会君
论文题目：物流中心选址与库存控制的双层规划模型及求解算法研究
作者简介：：孙会君，女， 1974年3月出生，2000年9月师从于北京交通大学高自友教授，于2003年11月获博士学位。

中  文  摘  要

物 流作为一个富有现代内涵的概念，被定义为“为了符合顾客所需要的必要条件，所发生的从生产地到销售地的物质、服务以及信息的流动过程，以及为使保管能有 效、低成本的进行而从事的计划、实施和控制行为”。随着世界经济的迅速发展和科学技术的不断进步，现代物流对经济贸易活动的影响与日俱增。特别是进入20世纪80年代以后，专业化分工进一步深化，对现代物流的研究也进一步细化，世界各国的物流研究也进入了一个崭新的阶段。
作 为与能流、信息流并列的物流业是二十一世纪的又一重要发展领域。在这个领域内，我国与发达国家不仅是资金、技术上的差距，更重要的是观念和知识上的差距。 现代物流需要打破地域间的阻隔和部门垄断，更需要有先进的物流技术。在众多的现代物流研究成果中，目前定性分析研究较多，而定量研究较少。本博士论文的主 要内容是运用优化方法的思想，特别是双层规划的思想和方法建立数学模型用以描述各种物流功能决策，通过设计合理的求解算法从理论上分析双层规划在物流系统 优化中的应用。其主要目的在于使物流决策部门或物流规划人员用尽可能少的资金，获得最佳的经济效果。为提高物流效率、降低物流成本、增加物流效益提供理论 基础与方法上的支持。
现 代物流系统中的物流中心有时也称为物流据点、流通中心、配送中心、集配中心等。它是现代物流系统的重要组成部分，其功能多种多样。物流中心的分布，对现代 物流活动有很大影响，科学地建立物流中心，是市场竞争的必然结果。物流中心合理的选址能够减少货物运输时间与费用，从而大幅度地降低运营成本，提高流通效 率。为了实现物流中心的合理分布，必须合理规划其布局，也就是要根据物流现状和预期发展，在特定条件下确定物流中心的地址。影响物流中心模式的最主要因素 包括：客户的数量、客户的地理分布、客户订单的大小、工厂及物流中心的数量和位置。在这四个主要影响因素中，供应商几乎无法控制前三者，而只能对最后一项 进行规划，同时也就意味着物流中心选址决策要确定建立物流中心的数量、位置及其服务区域等问题。
传 统的选址模型一般为单层混合整数规划问题，这些方法或者只考虑客户的需求量在某一个物流中心满足的情况，即每个客户只能由唯一的一个物流中心服务，或者只 考虑在总费用最小情况下的配送方案，而不考虑客户的选择行为。而在现代物流体系中，顾客的需要被放到了首位，在研究选址问题时也不应例外。事实上，某个客 户的需求不但可以由多个物流中心共同满足，而且由每个物流中心配送的货物量取决于客户的选择行为，同时这种选择行为在很大程度上具有较高的随机性。因此， 有必要以考虑客户选择行为的新模型、新方法、新技术来求解选址问题。
尽管双层规划理论已经应用到了许多领域，但在物流技术研究中相关的研究成果却不太多。Taniguchi（1999）介绍了在高速公路交叉口附近公共物流运输站点选址的双层规划模型，陆化普等（2001）也应用双层规划对物流中心的选址问题进行了描述，但它们均没有考虑市场竞争条件和路线安排等问题，且所用求解算法均为非数值优化方法，不能准确分析选址方案与客户需求之间的关系。同时由于双层规划是一个NP-hard问题，难以求解。因此，从理论上研究各种条件下物流中心双层规划选址模型和求解算法是十分有意义的。
我们可以把物流中心的选址问题看作一个领导者-跟 随者问题，其中决策部门是指导者，客户对物流中心的选择行为或者客户需求在各物流中心的分配为跟随者。决策部门可以通过政策和管理来改变某个物流中心的位 置和配送成本，从而影响客户对物流中心的选择，但不能控制他们的选择行为。客户则对现有的物流中心进行比较，根据自己的需求特点和行为习惯来选择物流中 心。这种关系我们可以用双层规划模型来进行描述。上层规划可以描述为决策部门在允许的固定投资范围内确定最佳的物流中心的地点以使得总成本最小（包括固定 成本和变动成本），此模型为一整数规划问题。而下层规划则描述了在多个物流中心存在的条件下，客户需求量在不同物流中心之间的分配模式，它的目标是使每个 客户的费用最低，即它满足用户最优的原则。模型考虑了客户的选择行为，是市场竞争条件下实际现象的合理反映。在不考虑已有物流中心存在的情况下建立了双层 选址模型，针对下层规划约束的特殊形式，设计了一种简单的启发式求解算法。
有 些不满载的配送任务从物流中心到客户都是采用巡回线路的，即若干个需求量较小的客户在一条配送线路上。因此，传统的费用表示忽视了对车辆巡回线路的考虑， 有可能造成对分销成本估计不准确。随着物质需求的多样性及贸易呈全球化趋势的发展，企业管理者希望协调好物流系统中的各个环节，即采用集成物流管理系统的 概念。因此，我们建立设施选址和运输路线安排综合考虑的双层规划模型，并设计了基于聚类的启发式求解算法。
上述模型假设在新物流中心建立前，不存在已有物流中心。本论文中又提出了考虑市场竞争的物流中心选址的双层规划模型（即需求可以在新旧物流中心之间分配）。这是一个非线性混合整数双层规划问题。以广义Benders分解算法中的支撑函数的概念为基本思想，利用上下层目标函数之间的关系设计了求解此模型的新算法。
从大跨度的研究角度出发，物流需求的分配遵循空间价格均衡规律，在考虑物流中心之间有相互影响的情况下，建立了基于空间价格均衡分配的双层规划模型，下层规划用一变分不等式描述，并用遗传算法对此模型进行了求解。
随 着生产社会化的进一步扩大，产品需求量也急速增加，客户要求更高的服务水平，对于一些公共物流中心如仓库等的经营者不得不考虑其改建或扩建问题，这类问题 也可以用双层规划进行求解，上层模型为确定最佳的投资方案，以使整个网络某种性能指标达到最优。下层模型为用户平衡配流模型。可以利用灵敏度分析的方法对 此双层规划进行求解。同时本论文还建立了需求为模糊情况下的扩建优化模型，将模糊机会约束转化为其清晰等价类后再按传统方法求解。
物 流中心的库存成本是其成本因素中很大的一部分，传统的经济批量模型及其它一些有关模型仅从供应方或需求方单方面考虑，没有同时考虑双方利益。而现在强调供 应链管理要求整个链上所有成员的共同利益最大。因此，在垄断市场环境中或当需求方与供应方建立了长期合作关系后，即需求方只在此供应商处订货的情况下，假 设供应商有优先决策权，我们建立双层规划模型来描述这种关系。上层规划从供应方的角度出发，确定最优的价格折扣，对需求方来说，通过每次的订购批量，一方 面可以获得价格上的优惠，另一方面，可以减少订购费用；对于供应方来说，虽然价格的数量折扣导致销售收入的减少，却可以使企业的订单处理费用及其它费用降 低。下层规划确定最优的订货批量，使需求方的成本最小。并应用基于差分的算法对模型进行了求解。
在 供应链二级分销网络中，网络规划决策部门（或上游供应商、各分布式工厂）决定生产产品的总量和库存量，以使其费用成本最低，分销商则根据生产商的产品价格 选择不同的工厂，这类问题也可以用双层规划描述。上层规划可以描述为决策部门在允许的生产能力限制范围内确定市场内各工厂最佳的产品生产量以使得总成本最 小。而下层规划则描述了在多个分销商存在的条件下，每个分销商的需求量在不同工厂之间的分配模式，它的目标是使供应链系统中所有分销商的总费用最低。同时 设计了简单的启发式算法对模型进行了求解。
对以上研究的各种优化模型及求解算法，本文还给出了一些相关的数值例子。
关键词：物流中心；选址问题；双层规划；扩建规模；库存控制；支撑函数；求解算法







Bi-level Programming Models and Algorithms for
Logistics Centers Location and Inventory Control
Sun huijun  
ABSTRACT
Logistics, as a modern concept, is defined as the process of planning, implementing and storage of raw material, in-process inventory, finished goods and related information from point of origin to point of consumption for the purpose of conforming to customer requirement. With the development of economy and technology, logistics system plays an important role on trade. Especially after 1980s’, with the deep of specialization, there comes to a new stage for the research of modern logistics.
Modern logistics, be paratactic to energy and information, is another developing field in 21 Centuries. In this field, compared with the developed counties, our country has the gap not only on capital and technology, but also on ideas and knowledge. In the numerous achievements about modern logistics, the qualitative analysis is much than the quantitative research. The main context of this doctoral dissertation is to develop models and reasonable algorithms in theory to describe the logistics function decisions with the thoughts of optimization, especially with the bi-level programming. The dissertation’s purpose lies in providing the theory and method supports in increasing efficiency and profits, decreasing logistics costs.
Logistics centers, as an important part in modern logistics systems, are also called as distribution centers or circulation centers and so on because of their different functions. The layout of logistics centers has a great influence on activities of the logistics systems. To build the scientific logistics centers is the result of market competition, and the reasonable location can reduce the cost and time of transportation and can decrease the operating cost and increase the circulation efficiency. In order to realize the proper distribution, it is necessary to design the logistics layout, that is to say it is needed to determine the location of logistics based on its current situations and developments. The main factors affecting logistics patterns include: the number, the distribution, the order of customers and the number and location of logistics centers. In the above factors, suppliers can not control the first three and only can control the last one. This means that the location decision is to determine the number, the location and the service areas of logistics centers.
The traditional location models, as single level ones, only either deal with the customers demand assigned in one logistics center or minimize the total cost, which don’t consider the choice behaviors of customers. In fact, the demand of a certain customer can be assigned among several logistics centers, and at the same time, the amount of the demand assigned in a logistics center depends on the choosing-behavior of customers, and this kind of behavior is very stochastic. Therefore, it is necessary to solve the location problem with new models, new method and new technology.
However, although the facility location problem has been studied widely, only very little attention has been given to model the location problem using bi-level programming model. Taniguchi developed a bi-level programming model to determine the optimal size and location of public logistics terminals. Lu also studied the bi-level model for logistics centers location problems. However, their research did not consider the marketing competition and the routing problem. And their solution algorithm was based on non-numerical methods which could not analyze the relationship between the location projects and the customers demand. Therefore, it is significant to study the bi-level model of location problems under several conditions in theory.
The location problem of logistics centers can be represented as a Leader-Follower problems or stackelberg game where the decision managers are the leaders, and the customers are the followers who choose the distribution centers. Therefore, it is suitable to represent the location problems with the bi-level programming models. It is assumed that the decision managers can influence, but cannot control the users’ choice behavior by changing the location patterns and the generalized costs through policies and managements. The customers compare and choose the existing logistics centers based on the habitual manner and demand characteristics. The upper-level of the bi-level programming model is to determine the optimal sites of logistics centers to make the total cost (fixed and variable cost) minimum and to meet the demands of customers at various locations, in the range of fixed investment formulated by decision makers. The lower-level model represents customers demand assignments among different logistics centers, and its objection is to minimize the cost of each customer. At the same time, assumed that there are not any old logistics centers before the new ones to be built. That is to say, we don not consider the competition between new and old logistics centers. In this case, the bi-level programming model is build, and we propose a simple heuristic algorithm in terms of the special form of lower-level constraints.
In most mathematical models for facility location it is implicitly assumed that the customers are served on a straight-and back basis so that the delivery cost pert unit can be expressed as a function of the radial distance between a logistics center and a customer site. In fact, if all customers on the route have individual requirement less than a truckload, then this case requires a visitation of customers/suppliers through tours. Therefore, the classical location problem ignores tours when locating facilities and subsequently may lead to increased distribution cost. With the variety of requirements, manager hope to coordinate all stages of logistics, that is to say integrating concept is adopted, which thinks that the facilities location, distribution and routing are dependent with each other. Therefore, in this paper, we use the bi-level programming model to describe these location problems. To estimate the costs of servicing customer from a location, we use hierarchical agglomerative clustering methods to estimate transportation cost.
The aforementioned models assume that before the new logistics centers are built, there are not any old ones. Therefore, another model considering the competition between old and new centers is proposed. This is a nonlinear mixed integral programming problem. Making use of the relationship between the upper and lower level problem, we adopt the concept of the support functions in Generalized Benders Decomposition (GBD) method to solve the model.
The distribution of logistics demand follows the spatial price equilibrium(SPE) rule for a large area. Under the condition of considering the interaction among different logistics centers, We establish the bi-level programming model based on the SPE and use the Genetic Algorithm to solve the model.
   With the development of economy, the products demand is increasing and the customers require higher service level. Therefore, the managers of logistics centers must rebuild or expand the logistics centers’ capacity. In general, to rebuild the centers is too cost, so the managers or government only expand the capacity of the existing logistics centers, which need invest some capital. Therefore, we solve this problem with the bi-level model too, the upper-level is to determine the optimal investments to make the system quality best and the lower-level is user equilibrium assignment model. Because the capacity can be treated as the continuous variables, we can make use of Sensitivity Analysis Based Method (SAB) and algorithm based on difference to solve the bi-level model. At the same time, expansion optimization model with fuzzy demand is proposed which can be transformed into the form of fixed value, so we can then solve it by traditional methods.
Inventory cost of logistics centers is an great percentage of the total cost, the traditional economy model and other relative models considered the problem only from the supplier side or from the purchaser side, and didn’t taken the both of them into account together. The Supply Chain Management emphasis on the common profit，therefore, we use the bi-level model to describe the relationship. The upper-level is to determine the optimal quantity discount from the points of suppliers. For the purchasers, on the one hand, they can get price discount; On the other hand, they can decrease the order cost, by order quantity each time. For the suppliers, although the price discount may reduce the sales income, it can lower the order process cost and other cost. The lower-level is to determinate the optimal order quantity and frequency to make the purchasers’ cost minimum. And the algorithm based on difference is given to solve the model.
   In the two-level distribution network of the supply chain, the network planners determine the outputs and inventories to minimize their cost, and the buyers choose different manufacturers based on their price of the products, which can be described by the bi-level programming too. The upper-level problem is that the planners decide the optimal outputs of the products within the production capacity and the lower-level model describes the distribution pattern of the demand to the total cost of the buyers.
We also give some numerical examples for abovementioned optimization models.
Key borad: Logistics centers; Location model; Bi-level programming; Expansion size; Inventory control; Support functions; Algorithms

afei2002

她的这篇博士论文已经出书——《现代物流与交通运输系统》～～～

richterzqx

我看应该属于管理科学/运筹领域

beyondmon

pfpf
名师出高徒

hxy1108

双层规划并没有什么神奇的地方，事实上，很多人以前就在用，只是没有从理论高度提高一下。国外很早就在搞了

方倩如

“针对下层规划约束的特殊形式，设计了一种简单的启发式求解算法。”“我们建立设施选址和运输路线安排综合考虑的双层规划模型，并设计了基于聚类的启发式求解算法。”“以广义Benders分解算法中的支撑函数的概念为基本思想，利用上下层目标函数之间的关系设计了求解此模型的新算法”。。。
我以前就差点栽进去了，后来才发现建立一个双层规划模型可能并不难，难的是怎样把模型求解并进行解释。孙肯定花的功夫不在少数，我很佩服她！！

lijigui

在国内在理论方面还是挺不容易的