基于离群修正CTGAN的船舶横摇阻尼智能预测模型  

Intelligent Prediction Model for Ship Roll Damping Based on Outlier-modified CTGAN

作　　者：刘悦 李敏 徐娜 丁军[1,2] 金建海 LIUYuee;LI Min;XU Na;DING Jun;JIN Jianhai(China Ship Scientific Research Center,Wuxi 214082,China;Taihu Lake Laboratory of Deepsea Technology and Science,Wuxi 214082,China)

机构地区：[1]中国船舶科学研究中心,无锡214082 [2]深海技术科学太湖实验室,无锡214082

出　　处：《中国造船》2024年第4期66-76,共11页Shipbuilding of China

基　　金：国家重点研发计划项目(2022YFB3306200)。

摘　　要：为了避免船舶在复杂的海洋环境中大幅横摇,提出了基于离群修正的条件表格生成对抗网络(conditional tabular generative adversarial networks,CTGAN)的船舶横摇阻尼智能预测模型。基于实际海洋观测和模拟收集的几何模型数据和船舶运动数据构建横摇阻尼数据集;用CTGAN学习数据分布,并改进数据分布的均衡性。提出了基于密度聚类的离群检测和修正算法,在低密度聚类簇区域利用样本点的互信息生成新样本点,降低生成数据中离群点对模型预测性能的影响。最后利用集成学习算法搭建横摇阻尼智能预测模型。消融试验结果表明,该模型的均方误差、平均绝对误差和决定系数分别为0.239,0.449和0.982,预测精度显著优于其他对比的模型,能为船舶横摇运动预报提供有用的信息。To prevent significant ship rolling in harsh marine environments,an intelligent prediction model for ship rolling damping is proposed based on outlier-modified CTGAN(conditional tabular generative adversarial networks).An RDD dataset was constructed using geometric model data and ship motion data collected from marine observations and simulations.Then CTGAN is employed to generate the rolling damping data by learning the distribution through adversarial training,thereby balancing and augmenting the quantity of the data.An outlier-modified algorithm based on DBSCAN is proposed.In low-density cluster regions,new samples are generated through interpolation using mutual information to reduce the impact of outliers on the model's performance.An ensemble learning algorithm is utilized to build the intelligent prediction model.Experimental results demonstrate that the model achieves significantly better performance than other comparative models with MSE(mean squared error),MAE(mean absolute error),and R2(coefficient of determination)being 0.239,0.449,and 0.982 respectively.The model can provide valuable information for predicting ship rolling motions and contributing to progress in ship engineering.

关 键 词：船舶横摇阻尼 智能预测 生成对抗网络 离群修正 集成学习 

分 类 号：TB126[理学—工程力学] TB183[理学—力学]