基于VMD-LSTM-MLR的短期电力负荷预测  

Short-term Power Load Forecasting Based on VMD-LSTM-MLR

作　　者：张震[1] 李孟洲 李浩方 马军强 ZHANG Zhen;LI Meng-zhou;LI Hao-fang;MA Jun-qiang(School of Electrical Engineering,Zhengzhou University,Zhengzhou 450001,China;Jiaozuo Power Supply Company,State Grid Henan Electric Power Company,Jiaozuo 454003,China;State Grid Henan DC Transportation Inspection Branch,Zhengzhou 450000,China;State Grid Zhejiang Shengzhou Power Supply Co.,Ltd.,Shaoxing 312499,China)

机构地区：[1]郑州大学电气工程学院,河南郑州450001 [2]国网河南省电力公司焦作供电公司,河南焦作454003 [3]国网河南直流运检分公司,河南郑州450000 [4]国网浙江嵊州市供电公司,浙江绍兴312499

出　　处：《水电能源科学》2021年第10期208-212,共5页Water Resources and Power

基　　金：国家重点研发计划(2018YFC0824XXX)。

摘　　要：针对短期电力负荷预测精度不高的问题,提出集合变分模态分解(VMD)、长短期记忆(LSTM)网络及多元线性回归(MLR)的VMD-LSTM-MLR预测方法。先使用VMD将电力负荷数据分解为特征、频率均不同的本征模态函数,然后用LSTM对高频部分进行预测,用MLR对低频部分进行预测,最后将LSTM与MLR得到的预测结果进行叠加,获得完整的预测结果。使用VMD-LSTM-MLR预测方法对江苏省某市电力负荷数据进行预测,验证了VMD-LSTM-MLR在预测电力负荷数据上有较高的精度,其平均绝对百分比误差MAPE、均方根误差RMSE均低于目前比较典型的改进算法,以及所列举的4种组合算法。Aiming at the low accuracy of short-term power load forecasting,the combination prediction method of variational mode decomposition(VMD),long short-term memory(LSTM)and multiple linear regression(MLR)was proposed.Firstly,the VMD was used to decompose power load data into eigen mode functions with different characteristics and frequencies.Then,the LSTM was used to predict the high frequency part,and the MLR was used to predict the low frequency part.Finally,the prediction results of LSTM and MLR were superimposed to obtain the complete prediction values.The VMD-LSTM-MLR method was used to forecast the power load data of a city in Jiangsu Province.It is verified that the VMD-LSTM-MLR has high accuracy in predicting power load,and its MAPE and RMSE are lower than that of the typical improved algorithms and the four combinatorial algorithms listed in this paper.

关 键 词：电力负荷预测 VMD LSTM MLR 本征模态函数 

分 类 号：TM721[电气工程—电力系统及自动化]