基于极大似然法的高速无人艇操纵响应模型参数辨识

高速水面无人艇操纵性预报精度取决于其运动模型中的参数获取精度，针对传统扩展卡尔曼滤波算法难以获取较高精度的模型参数问题，提出一种采用极大似然法辨识获取无人艇操纵运动2阶非线性响应模型参数的方法。基于某无人艇响应模型参数进行20° Z形仿真实验，采集艏向角和舵角变化数据，根据辨识原理与前向差分法设计一种极大似然辨识方法，通过辨识获取了模型参数。进一步研究发现在极大似然法辨识过程中部分参数有参数漂移现象产生，分析得到参数漂移产生的原因在于使用差分法处理Z形实验数据时忽略了舵角变化率的影响。采用正弦仿真实验数据结合极大似然法进行改进辨识研究，其舵角变化率可直接对舵角求导得到。针对极大似然法与扩展卡尔曼滤波算法的辨识结果展开操纵运动仿真实验。实验结果表明：通过极大似然法辨识获取的参数比传统卡尔曼滤波算法能更精确地预报无人艇的操纵运动，且基于正弦仿真实验数据辨识能有效解决极大似然法的参数漂移，从而为极大似然法辨识结果提供更高的精度。

Parameter Identification of High-speed USV Maneuvering Response Model Based on Maximum Likelihood Algorithm

The maneuverability prediction accuracy of high-speed unmanned surface vessel（USV）depends on the accuracy of parameter acquisition in its motion model. The high precision model parameters are difficultly obtained by commonly using the extended Kalman filter（EKF）method. The maximum likelihood（ML）method is used to identify the second-order nonlinear response model parameters of unmanned maneuvering motion. 20° zigzag simulation experiment is carried out to collect the data of heading angle and rudder angle with the parameters of an USV response model. A ML identification method is designed based on identification principle and the forward difference method, and the model parameters are obtained by identification. Further research finds that part of parameters identified by ML method are inaccurate because of parameter drift. The analysis shows that the reason of parameter drift is to neglect the influence of the rudder angle change rate for processing the zigzag experimental data by the difference method. An improved identification research based on ML method with sine simulation experimental data was carried out, in which the rudder angle change rate can be directly derived from the rudder angle. The simulation experiments of USV maneuverability motion based on the results identified by ML and EKF method were carried out. The experimental results show that the result identified by ML method is more accurate than that identified by EKF method, and the parameter drift can be solved effectively by identifying with sine simulation experimental data to improve the identification accuracy of ML method. Key