船舶冲击环境网络预报的参数主成分分析方法  

Principal component analysis method of parameters for network prediction of ship impact environment

作　　者：赵晓俊 郭君[1] 杨俊杰 赵华讯 ZHAO Xiaojun;GUO Jun;YANG Junjie;ZHAO Huaxun(College of Shipbuilding Engineering,Harbin Engineering University,Harbin 150001,China;Dalian Shipbuilding Industry Co.,Ltd.,Dalian 116005,China)

机构地区：[1]哈尔滨工程大学船舶工程学院,黑龙江哈尔滨150001 [2]大连船舶重工集团有限公司,辽宁大连116005

出　　处：《哈尔滨工程大学学报》2024年第9期1655-1661,共7页Journal of Harbin Engineering University

基　　金：国家科技重大专项(J2019-I-0017-0016).

摘　　要：针对由于船舶水下爆炸冲击的强非线性特征引起的在利用神经网络进行冲击环境预报时精度不高的问题,本文采用一种基于主成分分析的方法对网络模型的输入参数作降维处理从而提高精度。利用矩阵特征值提取和矩阵变换,通过主成分分析方法以及因子分析对原始数据样本进行数据降维处理,再选择适应的网络对冲击谱值进行快速预报。实验结果表明:主成分选取主要参考特征值的大小及下降趋势,保留陡降段的特征值,并分析过渡段特征值的取舍;同时验证了对参数实施去相关处理和降维处理可以明显改善神经网络的预报准确性。Aiming at the problem of low accuracy in shock environment prediction using neural networks because of the strong nonlinear characteristics of ship underwater explosion shock,a method based on principal component a-nalysis is used to improve accuracy by downscaling the input parameters of the network model.Using mathematical matrix eigenvalue extraction and matrix transformation,original data samples are subjected to dimensionality reduc-tion by principal component analysis and factor analysis.Then,the adapted network is selected for the fast forecas-ting of shock spectral values.The experimental results show that the selection of principal components mainly con-siders the size and decreasing trend of the eigenvalues,retains the eigenvalues of the steeply decreasing section,and analyzes the trade-offs of the eigenvalues of the transition section.Meanwhile,the implementation of the decor-relation and dimensionality reduction processing on the parameters can significantly improve the forecasting accuracy of the neural network.

关 键 词：参数降维 矩阵变换 因子分析 主成分 神经网络 水下爆炸 冲击环境 快速预报 

分 类 号：U661.44[交通运输工程—船舶及航道工程]