基于修正辅助变量法的船舶操纵响应模型辨识  

Identification of the ship maneuvering response model based on recursive refined instrumental variable least-squares

作　　者：曾道辉 蔡成涛 ZENG Daohui;CAI Chengtao(College of Intelligent Systems Science and Engineering,Harbin Engineering University,Harbin 150001,China;Key Laboratory of Intelligent Technology and Application of Marine Equipment,Ministry of Education,Harbin 150001,China;Modeling and Emulation in E-Government National Engineering Laboratory,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 [2]船海装备智能化技术与应用教育部重点实验室,黑龙江哈尔滨150001 [3]电子政务建模仿真国家工程实验室,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2023年第2期161-171,共11页Journal of Harbin Engineering University

基　　金：国家重点研发计划(2019YFE0105400);船舶总体性能创新研究开放基金项目(31422117)。

摘　　要：为了解决有色噪声干扰对船舶操纵响应模型参数的有偏估计问题,本文提出一种递推修正辅助变量最小二乘法对船舶操纵二阶响应模型进行参数辨识。算法在递推辅助变量最小二乘法的基础上,针对辨识模型引入动态滤波模型改善有色噪声的统计特性,使得系统被估参数值在有色噪声干扰的情况下也能无偏趋近于参数真值。为了验证所提算法的有效性,在实船Z形试验数据的基础上,将递推修正辅助变量最小二乘法、递推辅助变量最小二乘法、增广最小二乘法和标准最小二乘法所辨识得到的模型的进行比较。结果表明:递推修正辅助变量最小二乘法的辨识模型相比其余三者的辨识模型与船舶实际航向角的拟合精度和泛化精度更高,对应的均方根拟合误差和均方根泛化误差分别可达到1°和2°以下,最大绝对误差分别在3°和4°以下。To solve the problem of biased estimates of ship maneuvering response model parameters caused by colored noise,this paper presents a recursive refined instrumental variable least-squares algorithm to identify the parameters of the ship maneuvering second-order response model.Based on the recursive instrumental variable least-squares algorithm,the proposed algorithm introduces a dynamic filter to improve the statistical characteristics of colored noise,making the estimated parameter value closely approach the true value of the parameter with colored noise.To verify the effectiveness of the proposed algorithm,performance comparison experiments were performed on the recursive refined instrumental variable least-squares method(RRIVLS)algorithm,recursive instrumental variable least-squares method algorithm,recursive extended least-squares method algorithm,and recursive least-squares method algorithm based on the zigzag test data on a real ship.The results show the RRIVLS algorithm has a better fitting effect and generalization ability on the heading angle than the other three identification models.The corresponding fitting root mean square error and generalization root mean square error are below 1°and 2°,respectively,and the maximum absolute error is below 3°and 4°separately.

关 键 词：船舶操纵响应模型 参数辨识 有色噪声 辅助变量 最小二乘法 滤波模型 无偏估计 Z形试验 

分 类 号：U661.3[交通运输工程—船舶及航道工程]