基于LSTM循环神经网络的风力发电预测  

Wind power forecast based on LSTM cyclic neural network

作　　者：王炜 刘宏伟[2] 陈永杰 郑楠 李政[2] 纪项钟 于广亮 康健 Wang Wei;Liu Hongwei;Chen Yongjie;Zheng Nan;Li Zheng;Ji Xiangzhong;Yu Guangliang;Kang Jian(State Grid Shanxi Electric Power Company,Xi'an 710001,China;North China Electric Power University,Beijing 102200,China;North China University of Science and Technology,Tangshan 063000,China)

机构地区：[1]国网陕西省电力有限公司经济技术研究院,陕西西安710001 [2]华北电力大学,北京102200 [3]华北理工大学,河北唐山063000

出　　处：《可再生能源》2020年第9期1187-1191,共5页Renewable Energy Resources

基　　金：国网陕西省电力公司2019年科技项目(5226JY18000G)。

摘　　要：大规模风电接入电力系统会造成系统频率波动,利用不同高度的风速、风向的余弦值、温度、湿度、气压对风力发电数据进行准确预测,有利于制定合理的调度计划,降低风电对电力系统的影响。文章基于AGC自动发电控制的要求,选取每15 min为一个数据采集点,构建大数据集,建立了基于LSTM结构的循环神经网络超短期风力发电预测模型,并每15 min根据最新实际采集数据更新数据集,实现了预测网络的滚动更新。最后通过某风电场的实际数据进行验证,结果表明,该算法预测精度高,对超短期风力发电预测有良好的适用性。Large-scale wind power access to the power system will cause system frequency fluctuations.Using wind speeds at different altitudes,cosine values of wind direction,temperature,humidity,and air pressure to accurately predict wind power generation data is conducive to the development of a reasonable scheduling plan.Based on the requirements of AGC automatic power generation control,this paper selects a data collection point every 15 minutes,builds a large data set,and establishes a LSTM structure-based cyclic neural network ultra-short-term wind power generation prediction model,which is updated every 15 minutes according to the latest actual collected data.The data set implements a rolling update of the predictive network.Finally,the actual data of a certain wind field is verified.The verification results show that the algorithm has high prediction accuracy and good applicability to ultra-short-term wind power generation prediction.

关 键 词：风力发电 LSTM循环神经网络 滚动预测 超短期风力发电预测 

分 类 号：TK81[动力工程及工程热物理—流体机械及工程]