CEEMD双层分解和Granger因果变量选择风电功率预测  

Wind power prediction based on CEEMD and Granger causality variable selection

作　　者：刘勇[1] 杨熙卉 燕林滋 欧云[1] LIU Yong;YANG Xihui;YAN Linzi;OU Yun(Yinchuan Energy Institute,Yinchuan 750100,China;Ningxia Transportation School,Yinchuan 750200,China)

机构地区：[1]银川能源学院,宁夏银川750100 [2]宁夏交通学校,宁夏银川750200

出　　处：《中国测试》2023年第4期98-105,共8页China Measurement & Test

基　　金：宁夏自然科学基金项目(2021AAC03254);宁夏高等学校科学研究项目(NGY2020123)。

摘　　要：针对风电功率时间序列具有高度随机波动性而无法精准预测的问题,提出一种基于完备集成经验模态分解(CEEMD)、Granger因果关系检验和长短时记忆网络的新型混合预测方法用于预测风电功率。首先,为研究风电功率和风速的隐性相关性,通过CEEMD算法对风电功率和风速时间序列分别进行序列分解,实现双层分解。其次,通过Granger因果关系方法对各风速分量与各风电功率分量进行因果关系检验,分析风电功率分量与各风速分量间的相关性,以此实现各风电功率分量的输入变量选择。最后,采用长短时记忆网络对各风电功率分量进行预测,并集成得到最终的风电功率预测结果。通过风电厂的实际数据进行了试验,并与多个应用广泛的经典模型进行对比,结果表明该方法的预测精度取得了大幅度提高,能够对风电功率实现精准预测。Due to the high random fluctuation of wind power time series,a hybrid prediction method based on CEEMD,Granger causality test and LSTM was proposed to predict wind power.Firstly,the wind power and wind speed time series are decomposed respectively through CEEMD to achieve the two-layer decomposition.Secondly,the Granger causality method is used to test the causality between each wind speed component and each wind power component,so as to realize the selection of input variables.Finally,LSTM is used to predict each wind power component.The final predicted value of wind power is obtained through integration to build a complete wind power prediction model.The actual data of wind power plants were tested and compared with several widely used classical models.The results show that the prediction accuracy of this method has been greatly improved,and the accurate prediction of wind power can be realized.

关 键 词：风电功率预测 完备集成经验模态分解 长短时记忆网络 GRANGER因果关系检验 

分 类 号：TM614[电气工程—电力系统及自动化]