数据驱动的水下潜航器在线建模及运动控制技术

Data Driven Online Modeling and Motion Control of Autonomous Underwater Vehicle

摘要:

面向未来高机动水下潜航器对精确运动控制的需求，考虑到现有控制方法所面临的动态跟踪能力差、耦合系统控制精度低等不足，研究基于模型的控制方法。水下潜航器水动力非线性强、建模困难，研究了数据驱动的水下潜航器运动建模方法，可实现对潜航器未来运动的长时精确预报。面向微小型水下潜航器难以安装声学测速及推进器转速测量装置，研究了无声学传感信息下线速度多步预报技术，仅需惯导数据及推进器、舵角控制信号，实现对不同航次线速度长时精确预报。基于该建模方法，研究了水下潜航器 MPC 及基于模型的强化学习运动控制技术，基于仿真平台及无人机实物平台验证了控制方法的有效性。

关键词:水下潜航器;在线建模;模型预测控制;强化学习

Abstract:

For the control of underwater vehicle with high mobility，the present control methods have shortages，such as poor dynamic tracking ability and inaccuracy for coupling systems. Based on this situation，model based control method is greatly needed. The hydrodynamic model of underwater vehicle is difficult to model precisely due to its nonlinearity. For the efficient and accurate modeling of underwater vehicle’s motion，data driven modelling method is proposed. For the velocity prediction of micro vehicles with limited inner space，data driven modelling method is proposed. This method doesn’t need Doppler acoustic velocity measuring device. Only IMU data and control signals of rudder and thruster is needed. Based on this modelling method，MPC and model based reinforcement learning technique is studied and validated using simulation platform and UAV testing platform.

Key words:underwater vehicle;online modeling;model predictive control;reinforcement learning