基于正交小波和长短期记忆神经网络的用电负荷预测方法  

Electricity Load Forecasting Method based on Orthogonal Wavelet and Long Short-term Memory Neural Networks

作　　者：张林 赖向平 仲书勇 李柯沂 ZHANG Lin;LAI Xiangping;ZHONG Shuyong;LI Keyi(State Grid Chongqing Electric Power Company,Yuzhong District,Chongqing 400010,China;Chongqing Smart Power Grid Technology Co.,Ltd,Yubei District,Chongqing 401120,China;Information and Communication Branch of State Grid Chongqing Electric Power Company,Yuzhong District,Chongqing 400010,China)

机构地区：[1]国网重庆市电力公司,重庆市400010 [2]重庆智网科技有限公司信息通信分公司,重庆市401120 [3]国网重庆市电力公司信息通信分公司,重庆市400010

出　　处：《现代电力》2022年第1期72-79,共8页Modern Electric Power

基　　金：国家重点研发计划(2017YFB1401702);重庆市人工智能技术创新重大主题专项课题(cstc2017rgznzdyf0051)。

摘　　要：用电负荷数据的波动性和周期性会影响电力负荷预测的准确性,针对此问题,文中提出了一种基于正交小波长短期记忆神经网络(orthogonal wavelet transform-long short-term memory,OWT-LSTM)的用电负荷预测方法。该方法对用电负荷序列进行正交小波分解,消除负荷数据的波动性,然后利用LSTM及其变种神经网络对正交小波分解后的各尺度负荷序列进行建模训练,通过各序列预测结果进行预测重构,获得最终的负荷预测结果。通过用户用电负荷数据集验证表明,该方法的预测性能优于其他模型,具有较高的预测精确性和稳定性。Due to the fact that the volatility and periodicity generated by electrical load data affect the accuracy of power load forecasting,a electric load forecasting method based on Orthogonal Wavelet Transform-Long Short-Term Memory(abbr.OWT-LSTM)was proposed.Firstly,the orthogonal wavelet transform of electric load series was performed to eliminate the volatility of load data,then the LSTM neural network was utilized to conduct the modeling and training of all scales load series after orthogonal wavelet decomposition,and through the forecasting results of all series the forecasting results the prediction reconstruction was conducted to obtain final load forecasting results.Experimental results show that through the verification by consumers’load data set,the forecasting performance of the proposed method is evidently better than other models,and it possesses better forecasting accuracy an stability.

关 键 词：电力负荷 正交小波变换 长短期记忆网络 预测重构 预测精确度 

分 类 号：TM73[电气工程—电力系统及自动化]