基于STL模型的月售电量综合预测方法  

A Comprehensive Forecasting Method of Monthly Electricity Sales Based on STL Model

作　　者：刘莉[1] 王彦博 庞新富[1] 耿赫男 LIU Li;WANG Yan-bo;PANG Xin-fu;GENG He-nan(Key Laboratory of Energy Saving&Controlling in Power System of Liaoning Province,Shenyang Institute of Engineering,Shenyang 110136,China;State Grid Dandong Electric Power Supply Company,Dandong 118000,China;State Grid Zhangjiakou Electric Power Supply Company,Zhangjiakou 075000,China)

机构地区：[1]沈阳工程学院,辽宁省电网节能与控制重点实验室,辽宁沈阳110136 [2]国网丹东供电公司,辽宁丹东11800 [3]国网张家口供电公司,河北张家口075000

出　　处：《控制工程》2020年第11期1930-1936,共7页Control Engineering of China

基　　金：辽宁省高等学校国(境)外培养项目(2018LNGXGJWPY-YB034);沈阳市科技计划重点实验室专项(F16-091-1-00);国家自然科学基金资助项目(61773269);辽宁省高等学校创新人才支持计划(LR2019045);沈阳市中青年科技创新人才支持计划(RC190042)。

摘　　要：电力零售市场下的月售电量预测面向小规模用户的电力需求,相对于传统意义的负荷预测更易受季节和节假日因素的扰动。传统预测方法直接对电量序列建模预测并未考虑序列分量随时间变化规律,因此预测精度不高。本文提出一种基于STL模型的综合月售电量预测方法,首先利用STL模型特点设置季节分量变化率,针对季节拐点月份和非季节拐点月份的售电特性将其电量时间序列进行个性化分解,将影响月售电量的因素分解成季节分量、趋势分量和随机分量,然后考虑了3个分量随时间的变化特征,分别选取适当的模型进行预测,最后将各分量的预测值重构为月售电量的预测值。基于R语言编制了预测程序,并对某大学园区用电量数据进行案例分析,结果表明所提方法合理有效。In the electricity retail market,facing the demand for small users,the monthly electricity sales forecasting is more easily disturbed by random factors than the traditional load forecast.At the same time,the accuracy of monthly electricity sales forecasting needs to meet the new requirements of the assessment system.A comprehensive monthly electricity sales forecasting method is proposed based on STL(Seasonal and Trend decomposition using Loess)model.First of all,STL model features is used to set the change rate of season component.According to the sales characteristics of seasonal inflection point month and non-seasonal inflection point month,personalized decomposition for electricity sales time series is presented.The factors that affect the monthly sales quantity are decomposed into seasonal component,trend component and random component,then considering the time-varying characteristics of the three components,appropriate models are selected to predict them respectively.Finally,the predicted values of each component are reconstructed into the predicted values of monthly electricity sales.A prediction program is compiled based on the R language,and a case analysis,whose data is from a university campus power consumption,shows that the proposed method is reasonable and effective.

关 键 词：月售电量预测 售电特性 个性化分解 时间序列 STL模型 

分 类 号：TM76[电气工程—电力系统及自动化]