In recent years, the size of container ships has become larger, thus requiring a more evident assurance of the hull structural safety. In order to evaluate the structural safety in operation, it is necessary to grasp the encountered sea state. The aim of this study is to estimate the encountered sea state using machine learning from measurement data of ocean-going 14,000TEU container ships. In this paper, as a first step in the study, considerable amounts of virtual sea state data and corresponding ship motion and structural response data are prepared. A convolutional neural network (CNN) is developed using these data to estimate the directional wave spectrum of encountered sea based on the hull responses. The input parameters of the formulated CNN include the spectral values of ship motion and structural response spectrum. The output of the CNN includes the sea state parameters of the Ochi-Hubble spectrum, specifically, significant wave height, modal wave frequency, mean wave direction, kurtosis, and concentration of wave energy directional distribution. It is found from the performance examination that the developed CNN is capable of accurately estimating the sea state parameters, although the level of accuracy decreases when the hull response is low. However, the decrease in accuracy when the hull response is low has a weak influence on the evaluation of the structural response to the estimated sea state.

中文翻译：

---

使用卷积神经网络 (CNN) 的监测数据进行海况估计

近年来，集装箱船的尺寸越来越大，对船体结构安全的保障要求更加明显。为了评价结构在运行中的安全性，必须掌握所遇到的海况。本研究的目的是使用机器学习从远洋 14,000TEU 集装箱船的测量数据中估计遇到的海况。在本文中，作为研究的第一步，准备了大量的虚拟海况数据和相应的船舶运动和结构响应数据。使用这些数据开发了卷积神经网络 (CNN)，以根据船体响应估计遇到的海面的定向波谱。公式化的 CNN 的输入参数包括船舶运动的谱值和结构响应谱。CNN 的输出包括 Ochi-Hubble 频谱的海况参数，具体为有效波高、模态波频率、平均波向、峰度和波浪能量方向分布的集中度。从性能测试中发现，所开发的CNN能够准确估计海况参数，尽管当船体响应较低时准确度水平降低。然而，船体响应较低时精度的降低对估计海况下结构响应的评估影响很小。从性能测试中发现，所开发的CNN能够准确估计海况参数，尽管当船体响应较低时准确度水平降低。然而，船体响应较低时精度的降低对估计海况下结构响应的评估影响很小。从性能测试中发现，所开发的CNN能够准确估计海况参数，尽管当船体响应较低时准确度水平降低。然而，船体响应较低时精度的降低对估计海况下结构响应的评估影响很小。