基于注意力机制的CNN-LSTM风速预测模型研究  

Research on CNN-LSTM wind speed prediction model based on attention mechanism

作　　者：童奇 熊龙祥 王贯宇 涂佳黄[1,2] TONG Qi;XIONG Longxiang;WANG Guanyu;TU Jiahuang(College of Civil Engineering,Xiangtan University,Xiangtan 411105,China;Foshan University,Foshan 528225,China)

机构地区：[1]湘潭大学土木工程学院,湖南湘潭411105 [2]佛山大学,广东佛山528225

出　　处：《湘潭大学学报(自然科学版)》2025年第2期46-54,共9页Journal of Xiangtan University(Natural Science Edition)

基　　金：湖南省自然科学基金(2021JJ50027,2022JJ50038);湖南省教育厅项目(21A0103);广东省基础与应用基础研究项目(2022A1515240077);。

摘　　要：基于风力大小非线性、随机性和难以准确预测的特点,构建了以卷积神经网络(CNN)和长短期记忆神经网络(LSTM)为基础的短期局部风速预测模型,并采用TensorFlow深度学习平台进行模型参数调试.然后构建了一种基于注意力机制的CNN-LSTM-Attention风速预测组合模型,采用福建平潭岛风电场4个不同季节的典型日风速数据集为样本对该模型的预测精度进行测试.测试结果表明,在风速预测精度方面,CNN-LSTM-Attention模型优于CNN-LSTM模型和LSTM模型,特别是在风速剧烈变化的工况时,CNN-LSTM-Attention模型的预测精度提升更为显著,且预测结果的可靠性更高,这表明该模型对于不同的风速变化和不同的数据集具有更强的适应性和稳健性.Based on the characteristics that wind magnitude is nonlinear,stochastic and difficult to predict accurately,a short-term local wind speed prediction model based on Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)was constructed,and the model parameters were debugged using the TensorFlow deep learning platform for model parameter debugging.Then,a combined CNN-LSTM-Attention wind speed prediction model based on the attention mechanism is constructed,and the prediction accuracy of the model is tested by using the typical daily wind speed datasets of Pingtan Island wind farm in Fujian Province in four different seasons as samples.The test results show that the combined CNN-LSTM-Attention model outperforms the CNN-LSTM and LSTM models in terms of wind speed prediction accuracy,especially in the case of drastically changing wind speed conditions,the CNN-LSTM-Attention model improves the prediction accuracy more significantly and the reliability of the prediction results is more reliable,which suggests that the model can be more reliable for different wind speed variations and different datasets.wind speed variations and different datasets,which indicates that the model is more adaptable and robust to different wind speed variations and different datasets.

关 键 词：风速预测 长短期记忆神经网络 卷积神经网络 注意力机制 组合预测模型 

分 类 号：TK81[动力工程及工程热物理—流体机械及工程]