摘要
[目的]针对实际航行中船舶动态特性变化所导致的模型失准问题,提出一种基于误差监测机制的船舶操纵运动自适应在线建模方法。[方法]通过模型预测误差监测机制判断模型更新时机,结合滑动窗口技术和支持向量机,基于航行数据实现模型的自适应重训练更新。以KCS集装箱船为研究对象,在变航速的Z形和回转运动场景下对所提方法进行测试验证,并分析误差监测机制中的超参数选取对在线建模的影响。[结果]仿真结果表明,误差检测机制能够降低模型在线更新频率,节约计算资源。相较于离线方法,所提方法在船舶动态特性变化时能够及时更新模型,保障预测精度。[结论]所提方法适用于船舶自身属性、环境变化等引发的动力学特性变化场景,可为船舶运动在线建模与预报提供技术方法,具有实际的工程意义。
[Objective]Aiming to address the problem of model inaccuracy caused by ship dynamic changes during actual navigation,this study proposes an adaptive online modeling method for ship maneuvering motion based on an error monitoring mechanism.[Methods]The method determines model update timing through a model prediction error monitoring mechanism and realizes the adaptive retraining update of the model based on voyage data by combining the sliding window technique and support vector machine.Taking a KCS container ship as the research object,the method is tested and validated under zigzag maneuvering and turning circle motion scenarios with variable speed,and the influence of the error monitoring mechanism’s hyperparameter selection on the online modeling is analyzed.[Results]The simulation results show that the error detection mechanism can effectively reduce the frequency of online model updating and save computational resources.Compared with the offline method,this method can update the model in time when the dynamic characteristics of the ship change,thereby guaranteeing prediction accuracy.[Conclusion]The proposed method is applicable to scenarios in which the dynamic characteristics of ships change due to their own attributes,environmental changes,etc.Thus,it has practical engineering significance by providing a technical method for the online modeling and prediction of ship motion.
作者
余耀辉
刘溯扬
王子豪
谢文博
彭艳
YU Yaohui;LIU Suyang;WANG Zihao;XIE Wenbo;PENG Yan(Institute of Artificial Intelligence,Shanghai University,Shanghai 200444,China)
出处
《中国舰船研究》
北大核心
2025年第1期58-64,共7页
Chinese Journal of Ship Research
基金
国家自然科学基金资助项目(52101361)。
关键词
船舶
操纵性
人工智能
运动控制
运动预测
在线建模
系统辨识
时变系统
ships
maneuverability
artificial intelligence
motion control
motion prediction
online modeling
system identification
time-varying system
