一种基于内存计算的电力用户聚类分析方法

A method for cluster analysis of electric power consumers based on in-memory computing

随着智能电表与采集终端采集的用电数据迅猛增长,传统数据分析方法已经不能满足大数据环境下智能用电行为分析的需要。鉴于K-means算法具有计算效率高、容易并行化等特点,采用弹性分布式数据集与并行内存计算框架对其进行改进与并行化,减少作业的运行与输入输出操作时间,提高聚类分析的处理能力。对用电测量数据进行预处理构建实验数据集,实验结果表明本方法对电力用户聚类分析的准确率高于单机K-means方法,其处理速度和能力明显优于单机和基于Map Reduce并行计算框架的聚类方法,并对数据的增长具有较好的适应性。

With the rapid growth of electricity consumption data collected by smart electric meters and data acquisition terminals,the traditional data analysis method cannot meet the demand of smart power consumption behavior analysis in the big data environment. Since K-means algorithm demonstrates high calculation efficiency,easy parallelization and other characteristics,a method for improving and parallelizing K-means with the resilient distributed data set and parallel in-memory computing framework is presented,the running time of job operation and I / O operations is reduced,and the ability of clustering analysis is improved. The experimental data set is built by preprocessed electricity consumption data. Eexperimental results show that the accuracy rate by this cluster analysis method for electric power users is obviously better than the single machine K-means algorithm. The processing speed and ability of this method are superior to the single machine and the clustering method based on Map Reduce parallel computing framework,and this method has good adaptability for the growth of data.