【作者】 卢青；

【导师】 林锦国； 王捷；

【作者基本信息】 南京工业大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 在竞争、合作、动态的市场环境中，虚拟企业被认为是21世纪最有竞争力的企业运作模式。简单地说，虚拟企业就是由盟主企业联合其他资源互补的合作伙伴，为及时适应市场机遇而组建成的动态联盟。其中，盟主企业能否从众多的候选企业中选择出最优的合作伙伴是虚拟企业运作成功的关键一步。但目前人们还没有找到一种合理而又有效的方法来帮助盟主企业进行决策。因此，本文将从虚拟企业盟主的角度出发，率先大胆地尝试采用支持向量机算法来实现盟友的选择。此课题是省教育厅自然科学研究项目，编号：02KJB630001。 目前，在虚拟企业中主要有三种合作伙伴类型：供应商合作伙伴、生产商合作伙伴和销售商合作伙伴。本文根据指标体系的确定原则，分析了影响虚拟企业盟友选择的具体因素，然后确立了这三种不同类型的盟友选择综合指标体系。 统计学习理论是一种专门研究有限样本下机器学习规律的理论。它不仅考虑了对推广能力的要求，而且追求在现有有限信息的条件下得到最优结果。支持向量机是在统计学习理论基础上发展而来的一种新的模式识别方法，在解决有限样本，非线性及高维模式识别问题中表现出许多特有的优势。因而，本文的另一项主要工作是将支持向量机的理论和方法引入盟友选择领域。为此，我们还必须解决这三个问题：如何将盟友选择问题转化为分类问题、多类分类算法的选择以及怎样将分类结果转化为绝对分值。我们通过将各候选企业的对应数据拼接形成大向量，并定义候选企业之间关系的类别为“相对强”、“相对弱”以及“实力相当”，解决了将盟友选择问题转化为分类的问题。由于经典的支持向量机算法只给出了二类分类的算法，而在实际应用中，一般要解决的是多类识别问题。在具体比较了4种多类分类算法的基础上，本文选用一对一组合方式，采用3个二值分类的分类器来完成三类分类，解决了多类分类问题。最后将所有候选企业分别与其他所有候选企业进行类似循环赛的两两比较，并通过对类别标签的进一步定义和相应的数据处理，完成了核心为SVM算法的虚拟企业盟友选择流程。且本文通过大量实验，调整了多项式核函数的2个参数，选定参数性能最好的SVM完成系统的构建。最后作者还用了3个不同类型的实例，分别地证实了构建的基于SVM的虚拟企业盟友选择系统的有效性。更多还原

【Abstract】 Under the competitive, collaborative and dynamic market circumstances, virtual enterprise is known as most competitive management mode of 21 -century. In fact, virtual enterprise is that the predominant corporation associates with other mutually beneficial partners in order to take hold of market opportunities in time. And it is very crucial that the predominant corporation can correctly select the partner from lots of potential partners. But at present people has not found a reasonable and effective method to help the predominant enterprises making decisions. So in the paper we will boldly adopt Support Vector Machines (SVM) algorithm to achieve partner selection.There are three kind of partners for the moment principally, such as, suppliers, producers and sellers. According to some principles of index, we firstly consider specific factors of influence on virtual enterprise partner selection. Then we respectively establish the three kind of partner selection evaluating index.Support Vector Machines is a new and very promising classification technique. The approach is systematic and properly motivated by statistical learning theory. Training involves separating the classes with a surface that maximizes the margin between them. An interesting property of this approach is that it is an approximate implementation of the Structural Risk Minimization (SRM) induction principle. Thus we put the theory and method of Support Vector Machines to apply partner selection in this thesis. Therefore, we must solve three related issues. The fist issue is how to evaluate by classifying. We concatenate the vectors of each two potential partners to be a "big" vector. We can classify such "big" vectors into three types, namely "better", "equal" and "worse", based on what relation between the two potential partners is. Thus we can tell the relation between any two partners by classifying the "big" vector concatenate from the vectors of them. The second issue is multi-class classification algorithms of Support Vector Machines. The traditional Support Vector Machines only deal with the binary classification. In this paper, based on four types multi-class classification algorithm, we deal with 3-class classification by one against one method, in which three machines are built to distinguish any two classes respectively. The third issue is how to form absolute evaluations based on the results of classification. To provide absolute evaluations, we adopt a round-robin-like mechanism, in which each partner is compared with every other one, and receive a mark based on the result. Such marks are cumulated to get the final evaluation of the partner. We have done lotsof simulation experiments. We have tried different kernels and adjusted parameters to find the best fit for our problem and built our system on it. In the last, we respectively apply our system on three samples in order to confirm that the system is effective.更多还原