无监督学习的潜艇模型智能简化策略  

Intelligent Simplification Strategy of Submarine Model Based on Unsupervised Learning

作　　者：张伟 邱志刚 张焕清[2] ZHANG Wei;QIU Zhigang;ZHANG Huanqing(Wuhan Second Ship Design and Research Institute,Wuhan 430064,China;School of Automation,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉市第二船舶设计研究院,湖北武汉430064 [2]武汉理工大学自动化学院,湖北武汉430070

出　　处：《河南科技大学学报（自然科学版）》2019年第6期37-43,M0004,共8页Journal of Henan University of Science And Technology:Natural Science

基　　金：国家自然科学基金项目(51579201)

摘　　要：针对潜艇运动方程复杂、耦合性强,水动力系数测量周期长等问题,引入了水动力系数敏感性指数,量化水动力系数对潜艇姿态的影响程度。在简化传统直接阈值法的基础上,提出了基于无监督学习K-means聚类的阈值聚类法。利用9个特征参数简化潜艇运动方程。研究结果表明:在简化项数相同时,阈值聚类法的特征参数中,除战术回转直径外,其余8个特征参数的误差均小于或者等于直接阈值法,且阈值聚类法的最大误差为3.850%,最小误差仅为0.008 3%。Aiming at the problems of complex and strong coupling of submarine motion equation,and long measuring peried of hydrodynamic coefficients,the hydrodynamic coefficient sensitivity index was introduced to judge the influence of hydrodynamic coefficient on submarine attitude. According to simplied direct threshold method,the threshold clustering method based on unsupervised learning K-means clustering was proposed.Nine characteristic parameters was used to simplify submarine motion equation.The results show that with the same number of simplified items,for the characteristic parameters of threshold clustering method except tactical rotation diameter,the errors of other eight characteristic parameters are less than or equal to that of the direct threshold method. The maximum error of threshold clustering method is 3. 850%,and the minimum error is0.008 3%.

关 键 词：水动力系数敏感性指数 无监督学习 阈值聚类法 潜艇模型 

分 类 号：U661[交通运输工程—船舶及航道工程]