摘要
提出一种对船舶和海工结构非线性水动力响应进行长期预报的新方法,利用高斯过程回归(GPR)和贯序取样技术,建立机器主动学习模型,实现对非线性长期极值响应贡献度最大的少数海况的贯序识别,以及对GPR模型训练集的持续更新。通过对GPR模型的迭代训练,使长期极值响应的预测值快速向真实值收敛,避免在整个海况域内进行非线性时域水动力计算。采用所研发的机器学习模型,对两型大型集装箱船的非线性水动力长期响应极值进行预报,以验证模型的高效性和准确性;并通过敏感性分析,对初始训练集数据的选择给出建议。模型对非线性问题具有广泛适用性,可以在适应性改造的基础上解决船舶及海工结构其它非线性响应的预报问题。
The paper presents a new method for prediction of the long-term nonlinear hydrodynamic re-sponse of ships and offshore structures by combining the Gaussian process regression(GPR)with sequential sampling to develop a model of active machine learning.By using the new method,the few sea states having the most contribution to the long term extreme response were sequentially identified and continuously added to the initial training dataset of the GPR model.Through the iterative training of the GPR model,the predict-ed value of the long-term extreme response quickly converged to the true value,avoiding the nonlinear time-domain hydrodynamic analysis in the entire sea state domain.The developed machine learning model was used to predict the long-term extreme value of the nonlinear hydrodynamic response of two large container ships,verifying its efficiency and accuracy,and through sensitivity analysis,suggestions for training dataset selection were provided.The model has wide applicability to nonlinear problems,and can predict other non-linear responses of marine structures on the basis of adaptive modification.
作者
李林斌
李洛东
刘日明
LI Lin-bin;LI Luo-dong;LIU Ri-ming(China Classification Society,Beijing 100007)
出处
《船舶力学》
EI
CSCD
北大核心
2023年第9期1294-1303,共10页
Journal of Ship Mechanics
关键词
机器主动学习
高斯过程回归
贯序取样
非线性水动力响应
颤振
active machine learning
Gaussian process regression
sequential sampling
nonlinear hydrodynamic response
whipping
