【作者】 周明；

【导师】 王太勇；

【作者基本信息】 天津大学 ， 机械制造及其自动化， 2009， 博士

【摘要】 随着ERP、CRM以及MRP-II等管理信息系统在制造业企业中的广泛应用,企业积累了大量历史记录,但是缺乏有效组织、分析和集成信息的手段,无法为管理层提供辅助决策支持。而企业的采购工作要求管理者根据当前情况及时做出应对措施,因此更需要决策支持系统的帮助。本文针对我国制造业企业采购特点,详细分析了采购决策支持系统的需求。在传统DSS系统架构的基础上,结合先进的数据挖掘技术,提出了新型采购决策支持系统的结构框架,有机融合了DW、OLAP以及DM技术,既包括了传统的决策分析功能,又增加了多维数据展示和深层次的关联信息挖掘技术,为决策支持提供了多种方法。在详细分析数据仓库技术的基础上,通过概念模型设计、逻辑模型设计和物理模型设计实现了采购决策数据仓库,探讨了异构数据的整合方法。对采购决策的OLAP分析内容和方法进行了研究,在Analysis Services的基础上设计和实现销售、采购和供货等主题的多维分析,帮助企业直观准确地了解信息。针对关联规则增量更新维护,提出了改进的快速更新频繁模式(IFUFP)算法,处理最小支持度阈值不变的情况下,事务数据库增量更新的关联规则维护问题。算法根据项目在新插入事务记录和原数据库中的支持度,分为四种情况验证,将满足条件的节点插入IFUFP-tree,再调用频繁增长模式挖掘关联规则。IFUFP树中的父母节点和子女节点之间双向连结,加快了节点更新速度,最大程度地利用了已有挖掘结果,提高了决策支持系统的运行效率。针对增量挖掘问题提出了基于前缀树的频繁模式算法(IFP)。在扫描事务数据集时,将项目按照指定的规范次序添加到IFP树中,同时更新项目在项目头表中的计数值。当插入一定数量事务记录后,按照项目当前支持度降序排列项目头表,并按此顺序重构IFP树。完成后继续按照当前项目排序插入事务,并再次执行重构步骤。算法通过插入步骤和重构步骤的循环交替进行,一次扫描数据库就可以得到全部频繁项目集,满足了企业的实际需要。更多还原

【Abstract】 With wildly use of ERP, CRM and MRP-II systems in manufacturing enterprises, a large amount of historical transactions records have been accumulated in databases without analysis or integration. It is useless to the managers, not mentioned helping decision making. But the procurement department has to make quick response to the challenge market everyday. So it is necessary to build decision support system in order to help them making excellent choice with information extracted from historical transactions records.A new decision support system is introduced with integration of data warehouse, OLAP and data mining module which was developed by requirements analysis of manufacturing enterprises procurement. It can provide managers with traditional decision making functions, multi-dimensional data analysis and association rules mining from transaction records.A procurement data warehouse is built through conceptual model design, logical model design and physical model design. An OLAP analysis cube of sales, purchase and supply is realized following DTS and ETL by using Analysis services tools provided by MS SQL server which enable visual analysis for enterprises.For transaction databases usually grow over time and the association rules mined from them must be re-evaluated and some new association rules may be generated and some old ones may become invalid. An incremental IFUFP-tree maintenance algorithm is presented with modification of the FP-tree construction algorithm for efficiently handling new transactions. The original database and new inserted transactions are considered in four cases and the results are then put into the IFUFP thus efficient maintenance association rules can be achieved. Besides, the counts of the sorted frequent items are also kept in the header table which is the same as FP-tree algorithm. Bi-directional links will fasten the maintenance processes of association rules through which high proficiency DSS system can be reached.A novel tree structure, called IFP-tree (improved pattern tree), which captures database information with one database scan and provides the same mining performance as the FP-growth method. The IFP-tree introduces the concept of dynamic tree restructuring to generate a highly compact frequency-descending tree structure at runtime following insertion phase and restructuring phase. An efficient tree restructuring method that restructures a prefix-tree branch-by-branch is also proposed. Extensive experimental results show that the IFP-tree is efficient for incremental mining with a single database scan that improves the DSS function.更多还原