卷积神经网络在涌潮冲击桩柱体水动力特性上的应用研究
Study on the application of convolutional neural network in the hydrodynamic characteristics of piles under the impact of tidal bore
王旭 1屈科 2杨元平 3高榕泽 4门佳4
作者信息
- 1. 长沙理工大学 水利与环境工程学院,湖南 长沙 410114;水沙科学与水灾害防治湖南省重点实验室,湖南 长沙 410114
- 2. 长沙理工大学 水利与环境工程学院,湖南 长沙 410114;水沙科学与水灾害防治湖南省重点实验室,湖南 长沙 410114;洞庭湖水环境治理与生态修复湖南省重点实验室,湖南 长沙 410114
- 3. 浙江省水利河口研究院,浙江 杭州 310017
- 4. 长沙理工大学 水利与环境工程学院,湖南 长沙 410114
- 折叠
摘要
涌潮作为一种特殊的强非线性间断流,携带着巨大的能量.在涌潮区域,桥墩等桩柱式建筑物受到的涌潮冲击作用力巨大,可能对涉水结构造成严重破坏.因此,迅速预测和评估涌潮对桩柱体的冲击作用是预防涌潮破坏工作中的关键部分,对海洋工程建设和人民生命财产安全至关重要.以一维卷积神经网络(CONV1D)为基础,搭建涌潮对桩柱体冲击过程水动力特性演变的预测模型.同时利用开源CFD软件OpenFOAM和Wave2Foam库,建立涌潮数值水槽,通过数值计算得到涌潮对桩柱体冲击的水动力特性数据集.以多测点的水位时间序列为输入样本,完成涌潮对桩柱体冲击载荷的预测和评估.结果表明:该预测模型得到的涌潮对桩柱体的冲击载荷和数值计算结果基本吻合.在最大的误差平方和情况下,模型评估涌潮对桩柱体冲击作用的最大载荷误差平均值为3.69%,最大荷载到达时间误差平均值仅为2.11%;且只需要较少的计算资源便可获得较高的计算效率,能提前准确获取桩柱体所受载荷信息.通过卷积神经网络预测,可以评估灾害发生的可能性,提前采取防御措施,减少灾害造成的损失.
Abstract
As a unique form of strong nonlinear intermittent flow,tidal bore carries immense energy.In areas affected by tidal bore,pile structures such as bridge piers,characterized by pillar-like constructions,face significant impact forces,which could severely damage hydraulic structures.Therefore,prompt prediction and assessment of the impact of tidal bores on piles is a key component of tidal bore damage prevention,crucial for marine construction and the safety of people's lives and properties.Utilizing one-dimensional convolutional neural network(CONV1D),a prediction model for the evolution of hydrodynamic characteristics of piles during tidal bore impacts was developed.Moreover,with the aid of open-source CFD software OpenFOAM and the Wave2Foam library,a numerical tidal bore flume was established,generating a hydrodynamic characteristics dataset for bore tide impacts on piles through numerical calculations.Using multi-point water level time series as input samples,the model successfully predicts and assesses the impact load on piles caused by tidal bore.Results indicate that the predicted impact loads on piles by the model are in good agreement with the numerical calculations.Even in the case of the maximum sum of squared errors,the average maximum load error assessed by the model is 3.69%,with an average maximum load arrival time error of only 2.11%.Moreover,the model achieves high computational efficiency with minimal computational resources,allowing for early and accurate acquisition of load information on piles.Through neural network prediction,it is possible to assess the likelihood of disaster occurrence,take preventative measures in advance,and minimize damage caused by disasters.
关键词
涌潮/卷积神经网络/载荷预测/桩柱体/冲击载荷Key words
tidal bore/convolutional neural network/load prediction/pile structures/impact load引用本文复制引用
出版年
2024
NETL
NSTL
万方数据