摘要
针对船舶运营阶段水下辐射噪声不易直接测量的挑战,提出了一种面向局部数据的改进高斯混合模型,对少量的样本进行数据增强;通过对比多元线性回归模型、岭回归模型和Lasso模型对样本数据训练结果的影响,确定了适用于研究对象的Lasso代理模型,实现由近场水下辐射噪声到远场水下辐射噪声的预报;通过“蒙眼法”对代理模型的预报精度进行了交叉验证。结果表明:高斯混合模型中的成分分布对数据生成质量产生影响;改进后的高斯混合模型可在不预设高斯分布成分数量的前提下,填补样本空间中的稀疏区域;数据增强后的水下辐射噪声预报误差为1.9 dB(10~20000 Hz),较原始少量样本的水下辐射噪声预报精度提高了3 dB,全频段相关系数也得到进一步提升。该方法可有效支撑船舶在运营阶段对水下辐射噪声实时预报的需求。
To address the challenge of directly measuring underwater radiated noise(URN)from ships in navigation,this study proposes an improved Gaussian mixture model(GMM)for localized data pairs to enhance sparse hydroacoustic datasets.The performance of multiple regression models,including multiple linear regression,ridge regression,and Lasso models,was systematically compared using the augmented training data.Results indicate that the Lasso model exhibits optimal capability for predicting far-field URN from near-field measurements.Forecasting accuracy was validated through blindfold cross-validation.Key findings reveal that the component distribution in GMM significantly influences synthetic data quality,with the improved GMM effectively populating sparse regions in the sample space.The data-cnhanced surrogate model achieved a prediction error of 1.9 dB(10~20,000 Hz),representing a 3 dB improvement over conventional methods with limited samples,while demonstrating enhanced correlation coefficients.This methodology provides critical technical support for real-time URN forecasting of ships.
作者
叶林昌
刘媛
刘赟
张淼
李清
沈建平
YE Linchang;LIU Yuan;LIU Yun;ZHANG Miao;LI Qing;SHEN Jianping(National Engineering Research Center of Special Equipment and Power System for Ship and Marine Engineering,Shanghai 201108,China;Shanghai Marine Diesel Engine Research Institute,Shanghai 201108,China;School of Naval Architecture,Ocean and Civil Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《中国造船》
北大核心
2025年第2期133-145,共13页
Shipbuilding of China
关键词
水下辐射噪声
快速预报
代理模型
数据增强
高斯混合模型
underwater radiated noise
rapid forecast
surrogate model
data enhancement
Gaussian mixed model
