摘要
风电功率预测对电力系统的经济调度和运行至关重要。为了减少集合经验模式分解产生的高频本征模函数IMF1对预测结果造成的影响,使用小波包分解进一步将IMF1子序列分解成若干子系列。针对传统机器学习无法处理时间序列间关联信息和时间相关性的缺陷,提出了级联式卷积神经网络-门控循环单元预测模型,提取风电功率子序列、风速子序列和风向之间的耦合关系的隐含特征,并进一步挖掘各个时间序列之间的时间相关特征。实验结果表明,所提出的预测模型优于其他预测模型,并验证了所提预测模型的有效性。
Wind power prediction is critical to economic dispatch and operation of power system.In order to reduce the impact of the high-frequency intrinsic mode function IMF1 generated by ensemble empirical mode decomposition on prediction result,wavelet packet decomposition is used to further decompose IMF1 subsequence into several sub-series.Traditional machine learning method cannot deal with the correlation information and time correlation between time series.In view of this problem,this paper proposes a cascaded CNN-GRU prediction model to extract the implicit features of the coupling relationship between the wind power sub-series,the wind speed sub-series and the wind direction,and further explores the time-dependent features between the input time sub-series.Experimental results show that the proposed model is superior to other prediction models and effectiveness of the model is verified.
作者
殷豪
欧祖宏
陈德
孟安波
YIN Hao;OU Zuhong;CHEN De;MENG Anbo(School of Automation,Guangdong University of Technology,Guangzhou 510006,Guangdong Province,China)
出处
《电网技术》
EI
CSCD
北大核心
2020年第2期445-453,共9页
Power System Technology
基金
国家自然科学基金项目(61876040).
关键词
风电功率预测
二次模式分解
卷积神经网络
门控循环单元
wind power prediction
two-layer mode decomposition
convolutional neural network
gated recurrent unit
作者简介
殷豪(1972),女,副教授,研究方向为电力系统及其自动化,E-mail:3403446@qq.com;通信作者:欧祖宏(1996),男,硕士研究生,研究方向为人工智能算法在电力系统中的应用,E-mail:zuhong.ou@foxmail.com;陈德(1994),男,硕士研究生,研究方向为智能优化算法在新能源并网运行中的应用,E-mail:2205514025@qq.com;孟安波(1971),男,博士,教授,主要从事电力系统自动化、系统分析与集成等方面的研究工作,E-mail:menganbo@vip.sina.com。
