基于深度神经网络的群桩局部冲刷深度预测  

Local scour depth prediction of pile group based on deep neural network

作　　者：张丽萍[1] 梁丙臣[1] 张黎邦 张嶔[1] 杨博[1] ZHANG Liping;LIANG Bingchen;ZHANG Libang;ZHANG Qin;YANG Bo(Ocean University of China,Qingdao 266100,China;Shandong Provincial Key Laboratory of Marine Engineering,Qingdao 266100,China)

机构地区：[1]中国海洋大学,青岛266100 [2]山东省海洋工程重点实验室,青岛266100

出　　处：《水道港口》2023年第5期747-753,810,共8页Journal of Waterway and Harbor

基　　金：国家自然科学基金项目(51739010)。

摘　　要：群桩的局部冲刷研究对保障海洋结构基础稳定性至关重要,采用现行规范和经验公式所得到的群桩冲刷深度计算值偏于保守且过于离散。深度神经网络具有很强的非线性映射能力,利用神经网络和学者研究中的试验数据建立深度神经网络模型,进行冲刷深度的预测并对结果进行敏感性分析。分析表明,文章建立的模型冲刷深度预测值与实际试验值拟合效果良好,冲刷深度神经网络模型具有可行性和有效性,可为群桩支撑的海洋工程建筑的填埋深度及后期防护等提供理论依据,具有十分重要的工程意义和理论意义。The study of local scour of pile group is crucial to ensure the stability of marine structure foundation.The calculated values of scour depth of pile group obtained by using current codes and empirical formulas are too conservative and too discrete.The depth neural network has strong nonlinear mapping capability.A depth neural network model was established using the neural network and experimental data from previous studies to predict the scour depth and perform sensitivity analysis on the results.It is shown by the analysis that the predicted scour depth of the model established in this paper fits well with the actual test values,and the feasibility and effectiveness of the scour depth neural network model are demonstrated.The theoretical basis for the depth of landfill and later protection of the marine engineering buildings supported by pile groups can be provided,and significant engineering and theoretical significance are attached to it.

关 键 词：深度神经网络 群桩 冲刷深度 ANN预测 深度学习 TensorFlow应用 

分 类 号：U65[交通运输工程—港口、海岸及近海工程] U442[交通运输工程—船舶与海洋工程]