【作者】 赵媛；

【导师】 刘威； 白洪涛；

【作者基本信息】 吉林大学 ， 软件工程， 2012， 硕士

【摘要】 随着电力体制改革的深入，传统的供电企业面临越来越多的竞争。在这种竞争环境下，企业迫切需要建设全新的客户关系管理（Customer RelationshipManagement，CRM）系统，在市场的竞争中取得更大的主动权，进而提高供电企业的经济效益。供电企业现有的运营和支撑系统中积累了大量的客户信息和消费数据，有必要研究先进的数据仓库和数据挖掘技术，充分利用宝贵的数据资源，帮助企业对这些数据进行正确的分析，挖掘客户典型特征，从而揭示或验证数据中隐藏的商业规律和法则。本文从研究数据仓库、数据挖掘相关技术出发，结合供电企业的知识特点，阐述了聚类和决策树两种数据挖掘方法在供电CRM中的典型应用。本文组织如下：1.数据仓库与数据挖掘介绍了数据仓库、数据挖掘的基本概念和有代表性的若干数据挖掘方法如关联规则、决策树、聚类分析等，讨论其在供电客户分析中的潜在应用。2.数据准备及预处理根据数据挖掘的目的，将采集到的不同的数据信息清洗并转换为待分析的数据。3.建模1)客户细分模型分析客户相关数据，选择对客户价值影响大的客户属性，采用聚类K-Means算法，将客户进行细分，以支持企业针对不同客户群，提出不同的营销政策，有效改善企业的盈利状况。2)客户响应模型采用决策树SPRINT算法，根据客户自然信息、客户消费行为信息，构建客户响应模型，找到对特定业务感兴趣的用户特征。在构建决策树的过程中，针对连续型和离散型属性的特点，优化最佳分割点的选取，减少侯选分割点的数目，有效提高算法的效率。经过反复实验，选择合适的阈值，对已生成的决策树进行了预剪枝策略，提取出分类规则，找出用户的主要特征，从而指导业务人员进行有针对性的营销。更多还原

【Abstract】 The reform of the electricity system allows enterprises to participate in theelectricity market to competition. Under this environment, those companies shouldbuild the new customer relationship management, to impove competition power in themarket.The operating and support system of such enterprises have accumulated a greatdeal of customer consumpting data. Data Mining is an effective tool, taking fulladvantage of valuable data resources, to help us analyze these data correctly and findthe characteristics of customers.Detailed data mining technology including clustering and decision tree methodsare used in this paper, through combining the data warehose and data miningtechnologies and the characteristics of the field of power supply enterprisesknowledge.1. The introduction of Data mining technologyThe correlative basic concepts about data warehouse and data mining areintroducd in the beginning of this paper, such as association rules, decision trees,cluster analysis and others. After all,some possible applications of those methods arediscussed in the power supply enterprise.2. Data preparationBefore applying the data minging algorithems, the original data from differentsources are extracting、transforming and loading to pending request format.3. Modeling1) Customer Segmentation ModelA standard clustering algorithm K-Means is selected for analyzing customer data.For the different customer groups, enterprises can take the different marketingpolicies, to effectively improve the profitability of enterprises. 2) Customer Response ModelAccording to the customer natural information, and spending behaviorinformation, a customer response model is built through one of decision algoriths,SPRINT algorithm.To effectively improve the efficiency of this algorithm, the bestsplit point is optimized in the process of constructing a decision tree, aming at thecharacteristics of continuous and discrete properties. The appropriate threshold is alsoinvestigated for the decision tree generated pre-pruning strategy.更多还原