基于元学习的船舶操纵性预报模型研究  

Research on Ship Maneuverability Forecast Model Based on Meta-learning

作　　者：张泽瑞 董宜滔 陈伟民[1,2,3] 吴梓鑫 徐延军 ZHANG Zerui;DONG Yitao;CHEN Weimin;WU Zixin;XU Yanjun(Shanghai Ship and Shipping Research Institute Co.,Ltd.,Shanghai 200135;State Key Laboratory of Navigation and Safety Technology,Shanghai 200135;Key Laboratory of Marine Technology Ministry of Communications,Shanghai 200135;COSCO Shipping Technology Co.,Ltd.,Shanghai 200135)

机构地区：[1]上海船舶运输科学研究所有限公司,上海200135 [2]航运技术与安全国家重点试验室,上海200135 [3]航运技术交通行业重点试验室,上海200135 [4]中远海运科技股份有限公司,上海200135

出　　处：《舰船电子工程》2025年第1期166-169,196,共5页Ship Electronic Engineering

基　　金：国家重点研发计划(编号:2022YFB4300802)资助。

摘　　要：论文基于元学习提出了一种快速、精确的机器学习方法,通过对自由自航模操纵性试验数据分析与建模获得预报模型,进而用模型预报船舶操纵性能目标参数,最终对船舶设计工作起到先验作用。相较于依赖大数据集的传统机器学习方法,元学习可以在有限数据量的基础上建立精度较高且泛化性强的数据模型。首先,该研究利用基于模型的特征选择算法为各目标参数筛选与其相关性强的特征参数,然后对各目标参数建立不同的预报任务,最后利用元学习方法对各预报任务进行学习获得最优的预报模型。经测试,该文的元学习模型预报精度较高,12个目标参数预报任务的平均模型损失稳定在0.65,对于船舶设计前期操纵性能评估具有显著的工程意义。Based on meta-learning,this paper proposes a fast and accurate machine learning method,which obtains the forecast model with the free self-propelled model maneuverability experiment data.Using this model to forecast ship maneuvering performance's target parameters,it plays a priori role in ship design.Meta-learning is different from traditional machine learning methods that rely on large data sets,it can establish model with limited amount of data.Firstly,the model-based feature selection algorithm is used to screen strong correlation feature parameters for each target parameter.Secondly,different forecast tasks are established for each target parameter.Finally,the meta-learning method is used to learn each forecast task to obtain the optimal forecast model.After testing,the meta-learning model predicted accurately.The average model loss of 12 forecast tasks is stably at 0.65.This reasearch about the ship maneuvering forecast model has important engineering significance for manoeuvrability performance evaluation in the early ship designing stage.

关 键 词：船舶操纵性 元学习 预报模型 机器学习 

分 类 号：U661.33[交通运输工程—船舶及航道工程]