基于混合核函数ν-SVR的船舶操纵运动非参数建模方法  

Nonparametric modeling method of ship maneuvering motion based on theν-SVR with mixed kernel function

作　　者：姜立超 尚晓兵 王伟[1] 张智[1] 李嘉祺 JIANG Lichao;SHANG Xiaobing;WANG Wei;ZHANG Zhi;LI Jiaqi(College of Intelligent Systems Science and Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学智能科学与工程学院,黑龙江哈尔滨150001

出　　处：《智能系统学报》2024年第6期1376-1384,共9页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金项目(62303129);黑龙江省自然科学基金项目(LH2023F022);中央高校基本科研基金项目(3072024XX0404).

摘　　要：非参数建模方法已广泛应用于船舶操纵运动建模。本文提出一种基于混合核函数(mixed kernel function,MK)的ν-支持向量回归(ν-support vector regression,ν-SVR)非参数建模方法,并采用遗传算法(genetic algorithm,GA)优化超参数,即遗传算法优化的混合核函数ν-支持向量回归(genetic algorithm-mixed kernel-ν-support vector regression,GA-MK-ν-SVR)。为了提高ν-SVR的性能,提出一种混合核函数,该核函数结合径向基和多项式核函数来同时捕获全局和局部性能。基于遗传算法,模型的超参数实现了优化。采用SIMMAN2008提供的KVLCC2油轮自由自航测试数据对该建模方法的性能进行了评估,并与多种现有操纵模型的预测结果进行了对比研究。试验结果表明,所提出的GA-MK-ν-SVR模型对船舶操纵运动具有良好的预测精度和较强的泛化能力。Nonparametric modeling methods are widely used in modeling ship maneuvering motions.This study introduces a v-support vector regression(v-SVR)nonparametric modeling method based on a mixed kernel function(MK),optimized through a genetic algorithm(GA)called GA-MK-v-SVR.The MK aims to improve the performance of v-SVR by combining radial basis functions with polynomial kernel functions,thus capturing both global and local characteristics.A GA is employed to fine-tune the hyper-parameters.The performance of GA-MK-v-SVR was evaluated by using KVLCC2 free-running test data provided by SIMMAN 2008,and the results were compared with several maneuvering models.Experimental results show that the proposed GA-MK-v-SVR model delivers impressive prediction accuracy and robust generalization capabilities for ship maneuvering motion modeling.

关 键 词：非参数模型 混合核函数 系统辨识 ν-SVR 船舶操纵运动 遗传算法 SIMMAN2008 KVLCC2 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] U661.33[自动化与计算机技术—控制科学与工程]