电力客户需求高适配性关联抽取算法  

High adaptability association extraction method of power customer demand based on analytic hierarchy process and decision tree

作　　者：潘晖 赵岩 李麟 徐可 李景顺 PAN Hui;ZHAO Yan;LI Lin;XU Ke;LI Jingshun(Nanning Power Supply Bureau of Guangxi Power Grid Co.,Ltd,Nanning Guangxi 532000,China)

机构地区：[1]广西电网有限责任公司南宁供电局,广西南宁532000

出　　处：《太赫兹科学与电子信息学报》2023年第10期1257-1262,共6页Journal of Terahertz Science and Electronic Information Technology

摘　　要：为了准确、高效地分析电力客户需求,从而降低电力企业成本,提高电力服务的产品附加值,基于层次分析法,计算条件属性重要度,构建优先关系矩阵,结合模糊关系判断尺度,确定电力客户需求权重。度量决策树节点纯度,分别对离散型节点变量与连续型节点变量进行指标分析,判断电力客户需求权重的准确性。建立电力客户需求关联抽取模型,获取电力客户需求用户画像,将信息区分值作为区分变量能力强弱的指标,计算不同变量之间的相关系数,设计关联抽取算法,得到电力客户关联结果。该方法在高、中、低3种频率中,虽其平均绝对百分比误差(MAPE)值不断升高,且随着关联层次的增加而逐渐递增,但整体依旧较低,判断电力客户需求权重的准确性较高。In order to accurately and efficiently analyze the needs of power customers,thereby reducing the costs of power enterprises and increasing the added value of power service products,based on Analytic Hierarchy Process(AHP),the importance of conditional attributes is calculated,a priority relationship matrix is constructed,and the weight of power customer demand is determined by combining with fuzzy relationship judgment scales.The purity of decision tree nodes is measured and the indicator analysis is conducted on discrete and continuous node variables to determine the accuracy of power customer demand weights.A correlation extraction model is established for power customer demand,and a user profile is obtained.Taking the information differentiation values as the indicators of variable differentiation ability,the correlation coefficients between different variables are calculated.By designing correlation extraction algorithms,the power customer correlation results are obtained,and a user profile is got.Taking the information differentiation values as the indicators of variable differentiation ability,the correlation coefficients between different variables are calculated.By designing correlation extraction algorithms,the power customer correlation results are obtained.Among high,intermediate and low frequencies,the Mean Absolute Percentage Error(MAPE)values of this method are 87.3%,71.9%,and 54.1%,respectively.In intermediate-frequency customer data,the MAPE of this method is increased from 62.1%to 71.9%;in low-frequency customer data,MAPE is increased from 42.2%to 54.1%.This method has a good correlation effect.

关 键 词：层次分析 决策树算法 电力客户需求分析 高适配性 关联抽取 

分 类 号：TB497[一般工业技术]