首页|基于改进的多可信度高斯过程回归方法预测涡激振动问题

基于改进的多可信度高斯过程回归方法预测涡激振动问题

扫码查看
通过涡激振动快速预报工具DAVIV和多可信度数据融合方法,该文对涡激振动幅值响应进行了预测和重构.采用改进的多可信度高斯过程回归方法,结合大量低可信度DAVIV计算结果和少量高可信度实验数据,实现了单/多来流工况下,涡激振动幅值响应的分布重构和新来流工况下的幅值预测.同时,利用主动学习方法,在训练过程中自动增加新的数据点,获得最佳的数据采样情况,减少了参与训练的数据数量.在多来流工况下,综合考虑实验和数值计算预测过程中增加传感器数量或增加计算案例数的迭代策略,可更好地降低实际成本,为工程应用提供参考.
Prediction of Vortex-induced Vibration Problem Based on Improved Multi-fidelity Gaussian Process Regression Method
In this paper,the vortex-induced vibration(VIV)amplitude response is reconstructed and predicted with the VIV fast solving software which can be called DAVIV and the multi-fidelity data fusion method.The multi-fidelity method based on an improved nonlinear Gaussian progress regression combines a large number of low-fidelity DAVIV results with a small amount of high-fidelity experimental data.This method can be used to reconstruct the distribution of VIV response and predict the VIV amplitude under new oncoming flow conditions.Meanwhile,the active learning method is adopted to automatically add new data during the training process to obtain the optimal data sampling situation,which can reduce the data amount involved in training.For multiple oncoming flow conditions,the iterative strategy of increasing the number of sensors or increasing the number of analysis cases is considered during the experimental and CFD prediction process,which can better reduce the actual cost and provide a reference for practical engineering applications.

Vortex-induced vibrationMulti-fidelity methodData fusionGaussian progress regressionActive learning

徐立华、田中旭、范迪夏、王嘉松

展开 >

上海交通大学船舶海洋与建筑工程学院,上海 200240

上海海洋大学工程学院,上海 201306

西湖大学工学院,杭州 310024

涡激振动 多可信度方法 数据融合 高斯过程回归 主动学习

国家重点研发计划项目国家自然科学基金

2022YFC280630012172218

2024

水动力学研究与进展A辑
中国船舶科学研究中心

水动力学研究与进展A辑

CSTPCD北大核心
影响因子:0.594
ISSN:1000-4874
年,卷(期):2024.39(3)