文献类型：期刊文献

中文题名：基于多维特征分析的月用电量精准预测研究

英文题名：Research on the accurate forecasting of monthly electricity demand based on the multi-feature analysis

作者：唐静[1,2,3];李瑞轩[1];黄宇航[4];向万红[5];解来甲[5];彭一轩[5];宁立[5]

机构：[1]华中科技大学计算机科学与技术学院,湖北武汉430074;[2]广东海洋大学数学与计算机学院,广东湛江524025;[3]广东省大数据分析与处理重点实验室,广东广州510006;[4]国电江苏电力有限公司,江苏南京210036;[5]远光软件股份有限公司,广东珠海519085

年份：2017

卷号：45

期号：16

起止页码：145

中文期刊名：电力系统保护与控制

外文期刊名：Power System Protection and Control

收录：CSTPCD、、北大核心2014、Scopus、CSCD2017_2018、北大核心、CSCD

基金：广东省重大科技专项(2014B010117006);广东省大数据分析与处理重点实验室开放基金项目(2017005)

语种：中文

中文关键词：配用电大数据;用电量预测;多维特征分析;数据挖掘

外文关键词：big data in power distribution and consumption; electricity demand forecasting; multi-feature analysis; data mining

中文摘要：用户用电量的精准预测是智能配用电大数据应用和发展的关键之一。区别于传统的基于行业分类的预测办法,提出基于大数据挖掘技术的用户用电多维度特征识别,以及在此基础上的精准用电量预测方法。基于海量多用户用电特性,建立多维度用电特征评价指标体系。对用户用电特性空间进行聚类和分析,挖掘和识别用电模式。在不同的用电模式下,分别建立用电量时间序列预测模型,避免用电模式差异对预测算法准确性造成的不利影响。该方法适用于大数据平台的分析与处理,算例分析结果表明其相比以往方法能显著提高预测精度和稳定性。

外文摘要：Nowadays, accurate forecasting of electricity demand has become one of the most important technologies in the development and application of big data in smart power distribution and consumption system. Be different from traditional forecasting methods based on the classification of eclectic power industry, a new monthly electricity demand forecasting method which can be applied on today＇s typical big data platform is presented mainly through recognizing and utilizing the multi-dimension features of big data produced in the electricity consumption. Indexes to evaluate and classify multi-dimension features of electricity consumption are established through mass data in electricity consumption. Typical electricity consumption patterns in the user space are identified and obtained by data mining. To avoid the adverse effects of the discrepancy in consumption patterns on the accuracy of the prediction, different time series prediction models for different consumption patterns are established. The proposed method makes for the analysis and processing of big data platform, and the results show that the method can significantly improve the prediction accuracy and stability compared to previous method.