热带海洋学报 ›› 2024, Vol. 43 ›› Issue (4): 68-75.doi: 10.11978/2023111CSTR: 32234.14.2023111

• 海洋水文学 • 上一篇    下一篇

卷积神经网络方法在岛礁类海啸波水动力特性演变的应用

高榕泽1(), 屈科1,2,3(), 任兴月4, 王旭1,2   

  1. 1.长沙理工大学水利与环境工程学院, 湖南 长沙 410114
    2.洞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.水沙科学与水灾害防治湖南省重点实验室, 湖南 长沙 410114
    4.海南大学土木建筑工程学院, 海南 海口570228
  • 收稿日期:2023-08-05 修回日期:2023-09-04 出版日期:2024-07-10 发布日期:2024-07-22
  • 作者简介:

    高榕泽(1997—), 男, 山东平度人, 硕士研究生, 主要从事深度学习方面的研究。email:

  • 基金资助:
    国家重点研发计划课题(2022YFC3103601); 国家自然科学基金重点项目(51839002); 湖南省自然科学基金项目(2021JJ20043)

Application of convolutional neural network methods in the evolution of hydrodynamic characteristics of tsunamis like-wave over fringing reef

GAO Rongze1(), QU Ke1,2,3(), REN Xingyue4, WANG Xu1,2   

  1. 1. School of Hydraulic Engineering, Changsha University of Science & Technology, Changsha 410114, China
    2. Key Laboratory of Dongting Lake Aquatic Eco-Environmental Control and Restoration of Hunan Province, Changsha 410114, China
    3. Key Laboratory of Water-Sediment Sciences and Water Disaster Prevention of Hunan Province, Changsha 410114, China
    4. College of Civil Engineering and Architecture, Hainan University, Haikou 570228, China
  • Received:2023-08-05 Revised:2023-09-04 Online:2024-07-10 Published:2024-07-22
  • Supported by:
    National Key Research and Development Program of China(2022YFC3103601); National Natural Science Foundation of China(51839002); Natural Science Foundation of Hunan Province, China(2021JJ20043)

摘要:

海啸是严重的海洋灾害, 准确的海啸预测对于海洋工程和人民生命财产安全具有重要意义。本文以一维卷积神经网络(1-dimensional convolutional neural network, CONV1D)为基础, 构建岛礁地形的类海啸波水动力特性演变模型。通过输入类海啸波波高时程曲线的观测值, 得到岛礁指定地点的水位淹没时程曲线, 实现时间序列到时间序列的预测, 进行海洋灾害的实时预报, 提前布置防御措施以达到减小损失的目的。结果显示, 预测一组样本所需时间少于一秒, 相对于传统的地震海啸预警系统, 深度学习方法所需计算资源较少, 计算速度更快。对类海啸波到达时间预测的平均相对误差为0.71%, 最大水位高度预测的平均相对误差为6.99%, CONV1D得到的岛礁地形类海啸波水动力特性与数值结果吻合较好。

关键词: 深度学习, 卷积神经网络, 海啸预测, 水动力特性, 时间序列

Abstract:

Tsunami is a serious marine disaster, and accurate tsunami prediction is of great significance to marine engineering and the safety of people’s lives and property. In this paper, based on 1-dimensional convolutional neural network (CONV1D), the evolution model of tsunami-like hydrodynamic characteristics of reef topography is constructed. By inputting observed values of wave heights resembling tsunami waves, the water inundation time series curves for specified locations on islands and reefs are generated. This achieves a prediction from one time series to another, serving the purpose of marine disaster prevention. The results indicate that the average relative error in predicting the arrival time of tsunami-like waves is 0.71%, and the average relative error in predicting maximum water levels is 6.99%. The hydrodynamic characteristics of island and reef terrains resembling tsunami waves obtained through CONV1D exhibit a strong alignment with numerical results.

Key words: deep learning, convolutional neural network, tsunami prediction, hydrodynamic characteristics, times series