基于卷积门控循环单元的波浪发电系统输出功率预测  

OUTPUT POWER PREDICTION OF WAVE POWER GENERATION SYSTEM BASED ON CONVOLUTIONAL GATED CYCLIC UNIT

作　　者：吴凡曈 杨俊华[1] 杨梦丽 林炳骏 梁惠溉 邱达磊 Wu Fantong;Yang Junhua;Yang Mengli;Lin Bingjun;Liang Huigai;Qiu Dalei(School of Automation,Guangdong University of Technology,Guangzhou 510006,China;State Grid Henan Electric Power Company Ultra High Voltage Company,Zhengzhou 450052,China)

机构地区：[1]广东工业大学自动化学院,广州510006 [2]国网河南省电力公司超高压公司,郑州450052

出　　处：《太阳能学报》2024年第8期682-688,共7页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(62173148);广东省自然科学基金(2022A1515010184);广东省基础与应用基础研究基因项目(2022A1515240026)。

摘　　要：为高效准确预测波浪输出功率,提出卷积神经网络和门控循环单元混合模型波浪预测算法。采用间接预测方法,搭建直驱式波浪发电系统模型,运用CORREL函数分析不同波浪特征的相关性,结合卷积神经网络提取特征与高维空间中的波高关系,构造特征向量,通过门控循环单元网络进行训练,将全连接层的输出值经反归一化后获得预测波高值,输入所搭建模型,获得波浪输出功率预测值。仿真结果表明,与其他网络模型相比,在多特征输入情况下,混合模型波浪预测算法预测效率更高、精度更准确。In order to predict the wave output power efficiently and accurately,a mixture model of convolutional neural network and gated cyclic unit is proposed.The indirect prediction method is used to build a direct drive wave power generation system model,and CORREL function is used to analyze the correlation of different wave characteristics.Combining convolutional neural network to extract the relationship between characteristics and wave height in high-dimensional space,feature vectors are constructed.Through the gated cycle unit network for training,the output value of the full connection layer is inversely normalized to obtain the predicted wave height value.Input the built model to obtain the prediction value of wave output power.The simulation results show that the wave prediction algorithm of the mixture model is more efficient and accurate than that of other network models in the case of multiple feature inputs.

关 键 词：间接预测 波浪发电系统 卷积神经网络 门控循环单元 多特征输入 混合模型 

分 类 号：TM619[电气工程—电力系统及自动化]