Welcome to visit Acta Energiae Solaris Sinica,Today is Share:
ISSN 0254-0096 CN 11-2082/K

Acta Energiae Solaris Sinica ›› 2022, Vol. 43 ›› Issue (12): 495-502.DOI: 10.19912/j.0254-0096.tynxb.2021-0596

Previous Articles     Next Articles

COMPARATIVE STUDY OF SURROGATE CALCULATION MODELS FOR PERFORMANCE OF BLADE HYDROFOIL OF TIDAL TURBINE BASED ON DATA-DRIVEN

Hu Zhenhui1, Yuan Peng1,2, Si Xiancai1,2, Liu Yonghui1,2, Wang Shujie1,2, Zhou Ying1   

  1. 1. College of Engineering, Ocean University of China, Qingdao 266100, China;
    2. Ocean Engineering Key Laboratory of Qingdao, Qingdao 266100, China
  • Received:2021-05-31 Online:2022-12-28 Published:2023-06-28

数据驱动的潮流能水轮机叶片翼型性能保真替代计算模型对比研究

胡振辉1, 袁鹏1,2, 司先才1,2, 刘永辉1,2, 王树杰1,2, 周迎1   

  1. 1.中国海洋大学工程学院,青岛 266100;
    2.青岛市海洋可再生能源重点实验室,青岛 266100
  • 通讯作者: 袁 鹏(1975—),男,博士、副教授,主要从事海洋能及海洋机电装备方面的研究。yuanpeng50@hotmail.com
  • 基金资助:
    国家重点研发计划(2018YFB1501903); 山东省重点研发计划(2019GGX103012)

Abstract: After the improvement of the application mode of airfoil Bezier-PARSEC parameterization software FanOpt, the data set is established. Taking the Lift-to-Drag Ratio characteristics of airfoil as the target, the machine learning models such as support vector regression (SVR), decision tree, random forest regression, fully connected neural network and one-dimensional convolution neural network are used to fit the data, and the fitting accuracy of the training models is compared. The results show that the prediction accuracy of lift to drag ratio can reach 97.86% by using fully connected neural network and one-dimensional convolutional neural network as surrogate calculation models in the test set. However, compared with one-dimensional convolutional neural network, fully connected neural network has more advantages in dealing with this kind of data set with uncomplicated structure.

Key words: tidal energy, blades, machine learning, hydrofoils, parameterization, neural networks

摘要: 在对翼型Bezier-PARSEC参数化软件FanOpt的应用方式改进后建立数据集,以翼型升阻比特性为目标,分别利用支持向量回归(SVR)、决策树、随机森林回归、全连接神经网络、一维卷积神经网络等机器学习模型进行拟合,训练模型比较拟合精度。结果表明,全连接神经网络、一维卷积神经网络作为替代计算模型在测试集上对升阻比的预测准确率可达97.86%,但相比于一维卷积神经网络,全连接神经网络在处理这种结构不复杂的数据集时更有优势。

关键词: 潮流能, 叶片, 机器学习, 水轮机翼型, 参数化, 神经网络

CLC Number: