基于大数据的用户用电特性研究

发布时间：2018-09-18 18:23

【摘要】：近年来智能电网建设快速发展,与此同时建设了一套完善的电力用户用电信息采集系统,这一系统将为数据分析提供海量的原始数据。由于我国拥有众多的电力系统,所以数据的量级将达到PB级甚至TB级,传统的技术手段很难处理这个级别的数据,需要使用新兴的大数据技术来进行相关的分析。在当今社会数据是最重要的财富,数据中隐含着各种各样的信息,通过分析这些数据可以得到更有价值的信息,让这些数据发挥更大的作用。本文主要通过分析用户用电的相关数据,得出用户的相关用电特性,并对用户未来的用电数据进行负荷预测。首先,介绍了数据挖掘的相关概念,为数据的预处理提供了方法。面对海量的用户数据,本文介绍了常见的大数据处理框架Hadoop和Spark,同时搭建了相关的集群,为大数据的处理提供数据分析平台。然后将数据挖掘中的聚类技术应用到大数据分析平台之上,对用户的每日用电负荷进行聚类,得到用户的日负荷曲线并对负荷曲线进行相关的研究,得出不同用户的用电模式。由于用户的样本维度较高,传统的聚类方法的效果并不理想,本文在谱聚类的基础上进行相关的改进得到幂迭代聚类,将幂迭代聚类应用到数据分析中同时在Spark平台上进行了相关的实现,最后能够很好的得出相关用户的用电特性。最后,对用户进行负荷预测,将局部加权算法和Hadoop平台相结合实现负荷预测,并与真实的负荷进行对比,验证了该方法在电力大数据中的有效性与应用性。
[Abstract]:In recent years, the construction of smart grid has developed rapidly. At the same time, a set of perfect power user information collection system has been built, which will provide massive raw data for data analysis. Because China has a large number of power systems, the order of data will reach PB level or even TB level, the traditional technology is difficult to deal with this level of data, need to use the emerging big data technology to carry out related analysis. In today's society data is the most important wealth, there is a variety of information hidden in the data, through the analysis of these data can get more valuable information, so that these data play a greater role. Based on the analysis of the relevant data of the user's electricity consumption, this paper obtains the relevant characteristics of the user's electricity consumption, and forecasts the load of the user's future power consumption data. Firstly, the related concepts of data mining are introduced, which provides a method for data preprocessing. In the face of massive user data, this paper introduces the common big data processing framework Hadoop and Spark, at the same time set up a related cluster, which provides a data analysis platform for big data processing. Then, the clustering technology of data mining is applied to big data analysis platform to cluster the daily load of users, get the daily load curve of users and study the load curve. Because the user's sample dimension is high and the effect of the traditional clustering method is not ideal, this paper improves on the spectral clustering to obtain the power iteration clustering. The power iteration clustering is applied to the data analysis and implemented on the Spark platform. Finally, the power consumption characteristics of the relevant users can be well obtained. Finally, load forecasting is implemented by combining local weighting algorithm with Hadoop platform, and compared with real load, the validity and application of this method in power big data are verified.
【学位授予单位】：华北电力大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13;TM715

【参考文献】