【作者】 丁四波；

【导师】 张子刚； 黄卫来；

【作者基本信息】 华中科技大学 ， 管理科学与工程， 2008， 博士

【摘要】 近年来随着人们的保护环境和节约资源意识不断的增强,逆向物流越来越引起了政府、制造商和消费者广泛的重视,成为近年来的热点研究领域。逆向物流不仅节约资源、减少排放废弃物,降低企业成本,而且可以提高客户满意度,于是许多企业都纷纷建立自己的逆向物流体系。逆向物流网络是整个逆向物流体系运作的基础和关键。建立经济、高效的逆向物流网络是许多企业的目标,因此如何设计出合理的逆向物流网络在逆向物流领域具有实际的研究价值。论文首先介绍了选题的背景、意义和创新之处。现在对逆向物流的研究还处于起步阶段,逆向物流网络建模与优化有着广泛的应用背景。另外,本文分别从宏观和微观论述了选题的重要意义,提出了本研究的主要目标和内容。其次,论文总结了选址理论和差异演化算法(DE)的国内外研究现状,概述了基本选址模型。论文还指出了一些国内选址理论研究的特点和不足之处。对选址问题中经常使用启发式算法做了较为详细的介绍。第三,论文以排队论为基础研究逆向物流网络建模与优化。逆向物流中的处理回收物品的工厂被看作是服务台,回收物品被看作是顾客,而且回收的物品有不同的处理方法。本文先基于GI/G/1模型,为每个工厂从几个备选生产能力中选择一个建立逆向物流网络。为了求解该模型设计了遗传算法(GA)。第四,在GI/G/1模型的基础上将模型扩展为基于GI/G/m的逆向物流网络模型。随着求解问题的复杂性及难度的增加,提高GA的运行速度便显得尤为突出,并行遗传算法不仅提高了求解速度,而且由于种群规模的扩大和各子种群的隔离,使种群的多样性得以丰富和保持,减少了未成熟收敛的可能性,提高了求解质量。本文设计了并行遗传算法求解基于GI/G/m的网络优化模型。第五,在研究多层逆向物流网络建模与优化方面。考虑由居民、初级收集点、处理中心组成的多层逆向物流网络优化问题。在满足居民的需求的前提下,确定建立收集点和处理工厂的位置和数量。每个初级收集点有一个最大容量限制。本文建立了一个非线性整数规划模型,目标是最大化逆向物流收益。求解该模型用了自适应遗传算法。第六,在上面模型的基础上,把研究内容扩展为包括正向物流和逆向物流的网络结构。该模型要求同时优化正向和逆向物流网络。模型中生产销售商品的工厂和处理回收物品的工厂可以是同一个工厂,销售商店和初级回收点是不同的。销售商店和初级回收点以及工厂有容量限制的但是可以通过扩建来满足消费者的需求。本文设计了差异演化算法求解模型,该算法稳健性强、收敛速度快。最后,把多层逆向物流网络优化模型扩展为多期,多层,有容量限制,同时包括正向和逆向的网络优化模型。为了求解模型设计了模糊自适应差异演化算法。目前,模糊控制朝着自适应、自学习方向发展,使得模糊控制参数或规则在控制过程中自动地调整、修改和完善,从而使系统的控制性能不断改善,达到最佳的控制效果。通常差异演化算法采用固定的变异算子F和交叉算子CR,本文将模糊控制原理和差异演化算法结合,提出了一种模糊自适应差异演化算法,根据种群前后两代总体的差异和每个个体的差异,自动调整F和CR。更多还原

【Abstract】 Recently, the concept of protecting environment and saving resource prevail. Government, producer and customers pay a close attention to reverse logistics, and it has become a hot research field. There is a concern, in people’s mind, that the reverse logistics is not only the effective way to save resource, decrease waste, but also it can reduce the cost and improve the satisfaction degree of customers. Therefore, reverse logistics were built by many companies. The network of reverse logistics is the key of the whole reverse logistics. Many companies want to set up economic and high effective reverse logistics network, so how to built better reverse logistics network has practical value on research.Firstly, the thesis introduces the backgrounds, the significance, the innovative achievements and the motivations of choosing this topic. Although, the research on reverse logistics is justat beginning phase, but the design of network reverse logistics has huge applied fields. Also, this thesis, from macroscopic concept and microcosmic concept, presents the goal and main content of the research.Secondly, the thesis reviews the literatures of location theories and differential evolution algorithm, and summarizes the fundamental models of location theories. On the base of these, it analyzes the characteristics and insufficiencies of theoretic research of locating. Further more, the thesis introduce the heuristic search methods, such as greedy algorithm, genetic algorithm and differential evolution algorithm.Thirdly, based on the queuing theory, the writer studies the design of reverse logistics. The factories are regarded as service station, and returned goods are regarded as customers. There are many treatments to deal with the returned goods. Above all, this thesis, according to GI/G/1 model, selects production ability from many choices for a factory and write self-adaptive genetic algorithm to solve the model.Fourthly, this thesis extends the model based on GI/G/1 to GI/G/m model. Along with the complexity increasing, the running speed of genetic algorithm plays an important role. Parallel genetic algorithm (PGA) improves the speed of solving problem and, owing to the huge population and insulated sub-populations, enhances the diversity of population. So PGA can effectively prevent convergence of immature individual. The writer designs PGA to solve complicated optimizing model. Fifthly, the thesis studies the research on multi-echelon reverse logistics. Customers, collection points and factories form the multi-echelon reverse logistics network, and we select the site and quantity of the collection points and factories, after every customer’s satisfactory is met. Every collection point has the maximum capacity of collecting returned products. The model is nonlinear integer programming model and the goal of the model is to achieve the maximum revenue.Sixthly, the thesis extends the model to integrate and optimize the reverse logistics and positive supply chain, at the same time. The factories can be the same one to producs goods and deal with the returned product. But the departments and collections are different. The departments, collection points and factories can be extended to satisfy customers’ need. The differential evolution algorithm was designed to solve the model. To the optimization problem of nonconve, multi-modal and non-linear function, the algorithm is much more robust and quicker in convergence than other evolution algorithms.Finally, the thesis turns the model into a more complicated one. It is multi-period, multi-echelon, has capacity limitation and integrates the reverse logistics and positive logistics. In order to solve the model, a fuzzy self-adaptive differential evolution algorithm was developed. At present, fuzzy control develops towards adaptive control and self-learning control. The adaptive controllers modify the parameters and rules, so the best performance of the control system can be obtained, hi the classic differential evolution algorithm, the F and CR are fixed, the thesis, combining the fuzzy control and differential algorithm, develops a fuzzy self-adaptive differential evolution algorithm that modifies the mutation factor and crossover rate by using the difference of twe generations.更多还原