基于融合长短时记忆网络的风电场超短期风速预测研究  

Very-short-term wind speed prediction of multiple wind farms based on merged long-short term memory networks

作　　者：王红刚 李彬[1] Wang Honggang;Li Bin(Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学

出　　处：《可再生能源》2020年第1期41-46,共6页Renewable Energy Resources

基　　金：国家自然科学基金项目(61673305)

摘　　要：准确的风速预测是风电功率预测的重要基础,对于电力系统的安全、稳定和经济运行有着十分重要的意义。文章通过考虑临近风电场之间的风速时空相关性,提出了一种融合长短时记忆网络的多风电场超短期风速预测模型。首先,通过堆叠的长短时记忆网络提取单个风电场的时间序列特征。之后,通过张量拼接层以及全连接层融合多个风电场的时间序列特征。最后,使用线性全连接层输出所有风电场的未来风速预测值。采用江苏省3个临近风电场两年的数据来验证文章提出的模型。与4种常用方法的对比结果表明:融合长短时记忆网络在四个季节内的超短期风速预测结果均能达到最优;通过序列特征融合的方式可以考虑多个风电场之间的时空相关性。文章提出的时间序列特征提取和空间特征融合方案直观、有效,多个风电场的风速预测精度得到明显提升。Accurate wind speed prediction is an important basis for wind power prediction, and is of great significance to the security, stability and economic operation of power systems. In this paper, a very short-term prediction model based on merged long-short term memory network(MLSTM)is proposed by considering the spatial-temporal correlation of wind speed between adjacent wind farms. The model firstly extracts the time series features of a single wind farm through stacked long-short term memory networks, then merges the time series featured of multiple wind farms through the tensor concatenate layer and full-connection layer, and finally uses linear full-connection layer to output the predicted wind speed values of all wind farms. The model is validated by two-year data of three nearby wind farms in the south of China. Compared with four commonly used methods, the results show that the very short-term wind speed prediction results based on MLSTM can be better than the comparison methods in different seasons. The spatial and temporal correlation between multiple wind farms can be considered by the means of MLSTM to improve the accuracy of wind speed prediction.

关 键 词：风电 风速预测 长短时记忆网络 时空相关 

分 类 号：TK81[动力工程及工程热物理—流体机械及工程]