基于海量数据的销售预测研究与实现

发布时间：2019-04-28 19:05

【摘要】：随着近年来互联网的发展，企业已经拥有庞大的客户信息数据，这些数据的积累为企业提供了一种有效的营销途径。然而，企业累计的客户信息是非常庞大的，最初搭建的硬件设备不可能具备处理如此之大的海量数据的能力，仅仅是存储这些数据都是一笔巨大的开销。由于现有数据库系统的这些不足，导致了企业空有大量有用数据，却无法提取有用信息的尴尬处境。本文结合国内烟草企业面对不断剧增的业务数据，而现有的业务数据处理能力明显不足的现状，分析烟草企业构建Hadoop分布式数据处理平台的可行性，并详细介绍了Hadoop平台技术及其项目结构和体系结构。为满足市场需求，首先必须把握市场的实际需求，影响卷烟销量的市场因素是多样的。本文基于时间序列分解法预测模型，建立卷烟销售预测模型，并对模型进行了验证。具体研究内容包括以下几个部分：（1）针对目前烟草企业全国销售数据来源多、数据规模庞大等特点，，且基于企业现有数据库的实际情况，分析构建数据库营销系统的必要性，然后对该系统的总体设计目标和模块进行说明。（2）分析研究福建中烟营销平台的目前状况，依据实际需求，着重分析Hadoop在企业实际需求中可以胜任的数据处理技术，分析在烟草企业现有软硬件基础上构建Hadoop平台的的可行性。针对Hadoop平台中的关键技术HDFS和MapReduce做了深入研究，并以实例说明。（3）在分析Hadoop平台的的可行性之后，对各省市各规格卷烟到日的销售数据进行处理，建立销量预测模型，考虑到卷烟市场具有季节周期变化趋势和长期增长趋势的特点，建立符合卷烟市场特征的时间序列销量预测模型。该预测模型已经在企业中得到应用，指导企业生产和销售。
[Abstract]:With the development of Internet in recent years, the enterprise already has the huge customer information data, these data accumulation has provided a kind of effective marketing way for the enterprise. However, the accumulated customer information is very large, the initial hardware can not have the ability to deal with such a large amount of data, just to store the data is a huge overhead. Because of the deficiency of the existing database system, the enterprise has a lot of useful data, but can not extract the useful information. This paper analyzes the feasibility of constructing Hadoop distributed data processing platform for tobacco enterprises based on the situation that domestic tobacco enterprises face the increasing business data and the existing business data processing ability is obviously insufficient, and the feasibility of constructing the distributed data processing platform for tobacco enterprises is analyzed in this paper. The technology of Hadoop platform and its project structure and architecture are introduced in detail. In order to meet the market demand, first of all, we must grasp the actual demand of the market, and the market factors that affect cigarette sales are diverse. Based on the prediction model of time series decomposition, this paper sets up a model of cigarette sales forecast and validates the model. The specific research contents include the following parts: (1) in view of the current national sales data of tobacco enterprises from a large number of sources, data scale and other characteristics, and based on the actual situation of the existing database of the enterprise, The necessity of building a database marketing system is analyzed, and then the overall design objectives and modules of the system are explained. (2) analyze and study the current situation of Fujian Tobacco Marketing platform, according to the actual demand, emphatically analyze the data processing technology that Hadoop can be competent in the actual needs of enterprises. The feasibility of constructing Hadoop platform on the basis of existing hardware and software in tobacco enterprises is analyzed. The key technologies of Hadoop platform, HDFS and MapReduce, are studied in depth, and an example is given. (3) after analyzing the feasibility of Hadoop platform, the sales data of every province and city are processed, and the sales forecast model is established. Considering the characteristics of seasonal cycle and long-term growth trend of cigarette market, the cigarette market has the characteristics of seasonal cycle trend and long-term growth trend. The model of time series sales forecast is established which accords with the characteristics of cigarette market. The prediction model has been applied in enterprises to guide the production and sales of enterprises.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP311.13

【参考文献】