基于时间模式注意力机制的BiLSTM多风电机组超短期功率预测  

Ultra-short-term Power Prediction for BiLSTM Multi Wind Turbines Based on Temporal Pattern Attention

作　　者：王渝红[1] 史云翔 周旭 曾琦[1] 方飚 毕悦 WANG Yuhong;SHI Yunxiang;ZHOU Xu;ZENG Qi;FANG Biao;BI Yue(College of Electrical Engineering,Sichuan University,Chengdu 610065,China;State Grid Sichuan Comprehensive Energy Service Co.,Ltd.,Chengdu 610031,China)

机构地区：[1]四川大学电气工程学院,成都610065 [2]国网四川综合能源服务有限公司,成都610031

出　　处：《高电压技术》2022年第5期1884-1892,共9页High Voltage Engineering

基　　金：四川省科技计划(2021YFG0026)。

摘　　要：针对现有预测方法难以批量处理多风机间不同特征的问题,提出了基于时间模式注意力(temporal pattern attention,TPA)机制的双向长短时记忆(bidirectional long short-term memory,BiLSTM)网络多风电机组超短期功率预测方法。首先,基于集合经验模态分解(ensemble empirical mode decomposition,EEMD)获得风机原始功率信号的不同模态分量,以降低神经网络预测难度。其次,基于TPA机制,从Bi LSTM网络得到的隐藏行向量中提取多风机之间的复杂联系,从而使得具有不同特征的模态可以从不同时间步选择相关信息,进而降低各模态的预测误差。最后,将TPA机制与传统注意力机制应用于分散分布的14台风机区域功率预测任务。研究结果表明:基于本方法的多风电机组超短期功率预测的标准均方根误差仅为0.0546,证明TPA机制能有效提高多风电机组的超短期功率预测精度。Aiming at to solve the problem that the existing prediction methods are difficult to process the different characteristics of multiple wind turbines in batches,we put forward an ultra-short-term power prediction method for multi wind turbines based on bidirectional long short-term memory(Bi LSTM)network and temporal pattern attention(TPA)mechanism.Firstly,based on ensemble empirical mode decomposition(EEMD),different mode components of the original power signal of the wind turbine are obtained to reduce the difficulty of neural network prediction.Secondly,based on the TPA mechanism,the complex connection between multiple turbines is extracted from the hidden row vectors obtained from the BiLSTM network,so that the modes with different characteristics can select relevant information from different time steps,thereby reducing the prediction error of each mode.Finally,the TPA mechanism and the traditional attention mechanism are applied to the distributed power prediction task of 14 wind turbines.The results show that the normalized root mean squared error of the ultra-short-term power prediction for multi-wind turbines based on this method is only 0.0546,which proves that the TPA mechanism can effectively improve the ultra-short-term power prediction accuracy for multi-wind turbines.

关 键 词：超短期风电功率预测 多风电机组 时间模式注意力机制 双向长短时记忆 集合经验模态分解 

分 类 号：TM614[电气工程—电力系统及自动化] TP183[自动化与计算机技术—控制理论与控制工程]