【作者】 李桃迎；

【导师】 陈燕；

【作者基本信息】 大连海事大学 ， 管理科学与工程， 2010， 博士

【摘要】 随着信息化技术的发展,各领域系统中积累的数据越来越多,简单的查询统计功能已经满足不了实际需求,运用数据挖掘方法从现有数据中发现潜在、有意义的规律,获取有价值的知识,为高层管理与辅助决策提供依据已经成为解决问题的关键。因此,本文提出了“交通领域中的聚类分析方法研究”,主要包括：1.复杂多源异构数据整合方法研究,采用XML技术实现数据交换的接口,提供数据的共享与交互功能,解决各行业现行系统中数据的异构问题,从而满足不同系统之间数据的互联互通,为进行数据挖掘提供了数据准备。2.面向混合属性数据的权熵模糊c-均值优化方法研究,主要是针对现有算法的不足而提出。同时将其引入模糊关联规则中,以此提高了关联规则挖掘的精度和效率,同时拓展了模糊关联规则的应用范围。3.面向混合属性数据的聚类融合方法,提高聚类稳定性的同时,提高了聚类的精度和效率。给出了聚类融合的模型体系,并根据混合属性数据的特征进行了相应的扩充,包括分类、混合属性数据聚类成员的产生方法；共识函数的设计方法及步骤；簇的合并与分裂策略及步骤。4.研究基于聚类融合的混合属性数据增量聚类方法,针对增量聚类的研究中缺少对混合属性数据的研究,且增量方法易出现不稳定的现象,提出了基于聚类融合的增量聚类方法,分别讨论了有、无数据基础时的增量聚类问题,提高了聚类的精度和效率,节省了聚类的时间。5.研究聚类分析在交通领域中的应用,挖掘导致交通事故的原因和潜在规律,为相关管理部门提供辅助决策,预防交通事故的发生,确保国家、人民的生命财产安全。通过聚类在船舶等级划分中的应用,提高海事管理部门的管理效率,为管理者提供决策的依据。更多还原

【Abstract】 With the development of information technology, the data stored in database of all fields becomes more and more, and simple query and statistic methods are not enough now. Providing the proof for high management and assistant decision is the key of solving problem, which makes use of data mining for discovering potential and meaningful rules from existing data and obtains valuable knowledge. Therefore, the "Research on Clustering Algorithms and Their Application in Traffic Domain" is proposed in this dissertation, which can be shown as follows:1. Integration methods for complexity, isomerism and multiple sources data, this method adopts XML technology to implement the interface of data interchange and provides data sharing and exchange, and solves the problem of data isomerism among existing systems in any field. Then it can implement data interconnection and mutual communication and prepare the data for data mining.2. Weighted entropy fuzzy c-means optimization method for mixed numerical and categorical data, which is proposed for overcoming the disadvantages of existing algorithms. Then it is introduced into fuzzy association rules, which improves the accuracy and efficiency of association rule algorithm and broadens the application range of association rule.3. Study clustering ensemble algorithm for mixed numerical and categorical data, this algorithm is able to increase the stability, accuracy and efficiency of clustering. The structure of clustering ensemble models is given in this dissertation, and then we expands the models for mixed numerical and categorical data, including the methods of producing clustering memberships for categorical data and mixed data, algorithms and steps of designing integration functions, and merging and dividing strategies and its procedure.4. Incremental clustering algorithm for mixed numerical and categorical data based on clustering ensemble. The algorithm is proposed for solving problems that research on incremental clustering algorithms is little and existing incremental clustering algorithms is often unstable. Then the incremental clustering algorithms with history data and without history data are discussed respectively, which increase the accuracy and efficiency of clustering, and reduce the clustering time.5. Application of clustering analysis in traffic domain, it mines the reasons and potential rules leading to traffic accidents and aids decision making for related management departments, which can be used to prevent the occurrence of traffic accidents and guarantee the safety of the nation and people’s lives and property. The algorithm improves the management efficiency of maritime management organizations and provides proof of decision making by clustering applied in partitioning ship ranks.更多还原