As a kind of intelligent marine equipment, its successful recovery is the basis of Unmanned Underwater Vehicle (UUV)’s normal working and completing the mission successfully. However, the mathematical model of UUV is severely coupled and highly nonlinear, and may be subject to complex interference in the recovery process. Besides, the model could not be determined completely. Active disturbance rejection control (ADRC) method does not need the beforehand information of the unknown disturbance and also can ensure the stability. But conventional ADRC method with fixed parameters could not adjust to UUV complicated motion control. This paper introduces the adaptive wavelet neural network optimized by Levenberg-Marquardt algorithm, and puts forward a novel ADRC control method improved by wavelet neural network algorithm and Levenberg-Marquardt algorithm, for partial system identification and uncertain model compensation. Moreover, with pitching moment change considered in the process of UUV recovery, two dimensionless hydrodynamic coefficients are defined based on near-wall effect. The simulation experiments have been tested to verify the effectiveness of the proposed control. The results indicate that the ADRC with Levenberg-Marquardt neural network could control UUV recovery process in the variable disturbance environment more stable and reduce the ADRC computational burden.

作为一种智能船用设备，其成功的恢复是无人水下航行器（UUV）正常工作和成功完成任务的基础。但是，UUV的数学模型是严格耦合且高度非线性的，并且在恢复过程中可能会遇到复杂的干扰。此外，该模型无法完全确定。主动干扰抑制控制（ADRC）方法不需要事先知道未知干扰的信息，也可以确保稳定性。但是传统的固定参数ADRC方法无法适应UUV复杂的运动控制。本文介绍了通过Levenberg-Marquardt算法优化的自适应小波神经网络，提出了一种新的小波神经网络算法和Levenberg-Marquardt算法改进的ADRC控制方法，用于部分系统辨识和不确定模型补偿。此外，考虑到UUV恢复过程中的俯仰力矩变化，基于近壁效应定义了两个无量纲的水动力系数。仿真实验已经过测试，以验证所提出控制的有效性。结果表明，采用Levenberg-Marquardt神经网络的ADRC可以在可变扰动环境下更稳定地控制UUV恢复过程，并减轻ADRC的计算负担。基于近壁效应定义了两个无量纲的水动力系数。仿真实验已经过测试，以验证所提出控制的有效性。结果表明，采用Levenberg-Marquardt神经网络的ADRC可以在可变扰动环境下更稳定地控制UUV恢复过程，并减轻ADRC的计算负担。基于近壁效应定义了两个无量纲的水动力系数。仿真实验已经过测试，以验证所提出控制的有效性。结果表明，采用Levenberg-Marquardt神经网络的ADRC可以在可变扰动环境下更稳定地控制UUV恢复过程，并减轻ADRC的计算负担。