基于单参数自调节RM-GO-LSVR的船舶操纵灰箱辨识建模  

Grey box identification modeling for ship maneuverability based on single parameter self-adjustable RM-GO-LSVR

作　　者：梅斌 孙立成[1] 史国友[1,2] 马文耀[3] 王伟 MEI Bin;SUN Li-cheng;SHI Guo-you;MA Wen-yao;WANG Wei(Navigation College,Dalian Maritime University,Dalian 116026,Liaoning,China;Key Laboratory of Navigation Safety of Liaoning Province,Dalian Maritime University,Dalian 116026,Liaoning,China;Maritime College,Guangdong Ocean University,Zhanjiang 524088,Guangdong,China)

机构地区：[1]大连海事大学航海学院,辽宁大连116026 [2]大连海事大学辽宁省航海安全保障重点实验室,辽宁大连116026 [3]广东海洋大学海运学院,广东湛江524088

出　　处：《交通运输工程学报》2020年第2期88-99,共12页Journal of Traffic and Transportation Engineering

基　　金：国家自然科学基金项目(51579025);辽宁省自然科学基金项目(20170540090)。

摘　　要：为实现舵角小、试验数据少条件下船舶操纵辨识建模,提出了一种船舶操纵运动灰箱模型;搜集水动力系数已知的船舶运动数学模型作为备选参考模型(RM),计算被辨识船舶与备选RM的相关系数,并以此筛选合适的RM;运用相似准则将观测数据映射到RM的输入值域,建立被辨识船舶与RM的运动关联,获得了RM的加速度项,并使用线性支持向量回归(LSVR)机补偿被辨识船舶和RM加速度项间的误差;分析了机理模型,设计了合适的LSVR输入项,使用全局优化(GO)算法自动调节了LSVR的不敏感边界参数;基于自航模试验数据训练了灰箱模型,并与约束模试验(CMT)结果和计算流体力学结果比较,验证了灰箱模型的泛化能力和预报精度。研究结果表明:在20°船艏向、20°舵角Z形试验预报中,灰箱模型所得第一超越角精度至少比CMT、虚拟约束模试验(VCMT)和RM方法所得结果高1°,灰箱模型所得第二超越角精度至少比CMT和VCMT所得结果高0.4°;在35°舵角旋回试验预报中,灰箱模型所得进距精度至少比CMT、VCMT、数值循环水槽试验(NCWCT)和RM方法所得结果高1%,灰箱模型所得战术直径精度比CMT所得结果低4%,比NCWCT所得结果高10%;RM方法有助于灰箱辨识建模,GO算法能够优化LSVR的不敏感边界参数,建立的单参数自调节灰箱辩识建模方法能够实现小舵角、少数试验条件下的船舶操纵辨识建模。To realize the identification modeling of ship maneuverability when the rudder angle is small and the test data were less, a grey box model for the ship maneuverability motion was put forward. The mathematical ship motion models with foregone hydrodynamic coefficients were collected as the alternative reference models(RM). The correlation coefficients between the identified ship and the alternative RMs were calculated to select the appropriate RM. The similitude regulation was applied to map the measurement data to the input range of RM and to build the motion relationship between the identified ship and the RM, and the accelerations of RM were acquired. The linear support vector regression(LSVR) machine was used to compensate the acceleration error between the identified ship and the RM. The mechanism model was analyzed, the suitable LSVR inputs were designed, and the global optimization(GO) algorithm was used to automatically adjust the insensitive band parameter of LSVR. The grey box model was trained by the data of free running model test, and the results were compared with those of the captive model test(CMT) and the computational fluid dynamics to validate the generalization ability and prediction accuracy. Research result shows that for the zigzag test with 20° heading angle and 20° rudder angle, the prediction accuracy of the first overshoot angle from the grey box model is at least 1° higher than those of CMT, virtual captive model test(VCMT) and RM method. The prediction accuracy of the second overshoot angle from the grey box model is at least 0.4° higher than those of CMT and VCMT. For the turning circle test with 35° rudder angle, the prediction accuracy of advance from the grey box model is at least 1% higher than those of CMT, VCMT, numerical circulating water channel test(NCWCT) and RM method. The prediction accuracy of tactical diameter from the grey box model is 4% less than that of CMT, and is 10% higher than that of NCWCT. The RM method is benefited for the grey box modeling. The GO algor

关 键 词：船舶工程 船舶操纵 线性支持向量回归 辨识建模 验证试验 

分 类 号：U675.9[交通运输工程—船舶及航道工程]