基于分形特征提取的电能质量数据挖掘  

POWER QUALITY DATA MINING BASED ON FRACTAL FEATURE EXTRACTION

作　　者：于燕平[1,2] 林涛[2] 

机构地区：[1]柳州铁道职业技术学院动力技术学院,广西柳州545616 [2]武汉大学电气工程学院,湖北武汉430072

出　　处：《计算机应用与软件》2016年第11期24-27,65,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(51177111)

摘　　要：发现"海量"监测数据中电能质量问题,并提取出有用信息,是电能质量有效治理的关键。将大数据挖掘技术应用于电能质量知识发现,8类电能质量数据先进行EMD分解,取前2阶IMF参数后结合分形理论分别求取计盒维、截距等10维特征参数。经泛化处理后,训练并生成了可靠的决策树,抽取出IF-THEN分类规则,用于电能质量问题预测。通过对比分析,分形参数较其他特征参数更有利于如振荡暂态、切痕、尖峰、闪变等电能质量问题分析,特别是对含噪电能质量的分析。结合大数据挖掘技术,不含噪和含噪信号的平均识别率分别提高了1.8%和4.1%。To find the power quality problems from ' mass' monitoring data and to extract available information from it,this is the key to control the power quality effectively. We applied the big data mining technology to power quality knowledge discovery,exerted the empirical mode decomposition( EMD) on 8 kinds of power quality data in advance,took the first 2-order IMF parameters and then combined the fractal theory to calculate respectively 10-dimension feature vectors of box-counting dimension,intercept,etc. After the generalisation processing,they were trained and generated the reliable decision tree which was used to extract the classification rules of IF-THEN,and was used to forecast the power quality problems. By comparative analysis,the fractal parameters were more conducive to analysing the power quality problems such as oscillatory transient,notch,spike,flicker etc.,than other features,especially to the analysis of power quality signals with noise. Combining the big data mining technology,the average recognition rate of signals with and without noise increased 1. 8% and 4. 1% respectively.

关 键 词：数据挖掘 EMD分解 分形理论 决策树 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]