不同正则化方法在船舶辐射噪声计算中对比与试验研究

Comparison of different regularization methods in ship-radiated noise evaluation and experimental research

为研究不同背景下不同正则化参数方法在船舶辐射噪声评估中的适用性,降低辐射噪声计算误差,介绍了基于工况传递路径分析(OTPA)的船舶辐射噪声计算方法,讨论了奇异值分解(SVD)法、基于L-curve和广义交叉验证(GCV)的Tikhonov正则化方法在辐射噪声传递函数求解问题上的适用性,并进一步提出基于准最优(Q-O)判别准则的Tikhonov法。千岛湖舱段模型水下声辐射试验结果表明:在不同环境下,SVD方法性能最差;在正常环境下,L-curve、GCV和Q-O准则方法性能基本相同;随着背景噪声水平的升高,L-curve方法优于GCV方法,而Q-O准则方法优于L-curve方法,可较L-curve降低辐射噪声误差约0.42 dB,较SVD方法降低约1.19 dB。同时,Q-O准则鲁棒性分析表明该参数可近似视为最佳正则化参数,该方法可为后续实船工程应用提供指导。

In order to study the applicability of different regularization parameter methods in ship-radiated noise evaluation under various noise levels,and reduce the evaluated errors,the paper systematically introduces a method of ship-radiated noise evaluation based on the operational transfer path analysis(OTPA),and discusses the applicability of the singular value decomposition(SVD)method,Tikhonov regularization based on L-curve and generalized cross-validation(GCV)in solving the inverse problem of the transfer function.Then,the Tikhonov method based on quasi-optimality(Q-O)criterion was proposed to reduce the errors.Experiments with a reduced-scale cabin in Thousand Islet Lake show that the performance of the SVD method is the worst under various noise levels,and the performance of L-curve,GCV and Q-O criterion methods are almost the same.However,with the noise levels increasing,the L-curve method is better than GCV;and the Q-O criterion method is better than L-curve,which can reduce errors by up to 0.42 dB compared with the L-curve method and 1.19 dB compared with the SVD method.Meanwhile,the robustness analysis of Q-O criterion shows that the parameter can be regarded as the quasi-optimal regularization parameter,which can provide guidance for engineering application of ship-radiated noise evaluation.