基于深度学习的溃坝涌波爬高过程研究  

Study on the dam-break wave induced runup process based on deep learning

作　　者：陈家祺 曾俊 刘海江[1,2] Jia-qi Chen;Jun Zeng;Hai-jiang Liu(Colleague of Civil Engineering and Architecture,Zhejiang University,Hangzhou 310058,China;State Key Laboratory of Coastal and Offshore Engineering,Dalian University of Technology,Liaoning 116024,China)

机构地区：[1]浙江大学建筑工程学院,杭州310058 [2]大连理工大学海岸和近海工程国家重点实验室,大连116024

出　　处：《水动力学研究与进展（A辑）》2022年第3期426-431,共6页Chinese Journal of Hydrodynamics

基　　金：大连理工大学海岸和近海工程国家重点实验室开放基金(LP20V1)。

摘　　要：该文将卷积神经网络应用于实验室内的溃坝涌波爬高图像数据集,通过对比FCN和U-Net两类语义分割网络,遴选出U-Net作为深度学习模型展开应用。结果表明,U-Net模型在溃坝涌波爬高识别任务中能达到令人满意的效果,但其迁移能力还有待进一步提升。由于该模型对湿润床面上所记录图像中干扰信号的抵抗能力较弱,故对于未参与模型训练的溃坝涌波水体回流图像的识别效果较差。此外,该文还对模型深度学习过程中的三个超参数(学习率、批尺寸和卷积核尺寸)开展了敏感性分析,揭示了这三个超参数对模型预测结果的影响,为后续研究提供了相关参考。In this paper,the convolutional neural network is applied to study the dam-break wave induced runup process in the laboratory experiment.After comparing the performance between the FCN and U-Net,the U-Net model is selected as the present deep leaning approach.Results show that the U-Net model is effective in recognizing the dam-break wave run-up process,whereas the migration ability of the model performance is relatively poor.Model performance is also less satisfactory under the circumstance of the wet bed surface because there exist substantial interferences and difficulties in image recognition.With respect to the images recorded during the backwash phase,the model learning accuracy is poor,ascribed to the fact that these images are not fed into the model training process.In addition,a sensitive analysis is carried out in terms of the three model hyperparameters,i.e.,learning rate,batch size,and convolution kernel size.Impacts of these three hyperparameters on the performance of the model is revealed,which is helpful for the subsequent studies.

关 键 词：物理试验 卷积神经网络 溃坝涌波 爬高 超参数 

分 类 号：P753[交通运输工程—港口、海岸及近海工程]