【作者】 杨斌；

【导师】 桑宏强；

【作者基本信息】 天津工业大学 ， 机械工程（专业学位）， 2025， 硕士

【摘要】 波浪滑翔器是一种海洋无人自主航行器,在海洋探测、环境监测等领域具有广泛应用。路径跟踪控制是波浪滑翔器运动控制的重要技术,良好的路径跟踪精度更是波浪滑翔器完成海上自主航行任务的关键。然而,波浪滑翔器依靠环境驱动、易受环境干扰,在应用传统路径跟踪控制系统时可能会产生较大路径偏差。强化学习拥有环境交互式学习的特点,可被用于减弱外部干扰对无人艇路径跟踪控制系统的影响。鉴于此,本文研究基于强化学习的路径跟踪控制方法,以提高波浪滑翔器的路径跟踪精度和在不确定海洋环境中的适应能力,主要工作内容如下:首先,总结波浪滑翔器的国内外研究现状及其路径跟踪控制的研究进展,结合强化学习在无人艇路径跟踪控制中的应用现状,探讨强化学习对波浪滑翔器路径跟踪控制的重要意义。同时构建波浪滑翔器动力学模型并开发配套岸基监控系统,为仿真与海试建立基础。其次,设计适用于波浪滑翔器的航向控制器。利用近端策略优化算法对比例、积分和微分参数进行动态调节,以提升航向控制器在不同工况下的控制性能。仿真结果表明,与传统粒子群算法和反向传播神经网络相比,基于近端策略优化的方法能使控制器在保证航向跟踪精度的前提下更好地克服外部干扰。再次,在传统视线制导方法的基础上,提出一种基于近端策略优化的时变前视距离方法。该方法通过智能体实时感知路径跟踪偏差相关信息,自适应地调整前视距离,优化视线制导律的动态性能,从而提高波浪滑翔器的路径跟踪精度。仿真结果表明,该方法能够在不同路径、不同海况条件下有效降低航迹偏差,改善路径跟踪效果。最后,在仿真验证的基础上,通过海试验证所提出方法在真实海洋环境中的可行性与有效性。试验结果表明,所设计的路径跟踪控制系统能够有效提高波浪滑翔器的路径跟踪精度,增强其在海洋环境中的适应性,为波浪滑翔器的进一步研究与应用提供指导。更多还原

【Abstract】 The wave glider is an unmanned autonomous underwater vehicle,which has a wide range of applications in the fields of ocean exploration and environmental monitoring.Path-following control is a crucial technology for the wave glider motion control,and its accuracy directly affects the execution of autonomous maritime missions.However,due to their reliance on environmental forces and susceptibility to disturbances,conventional path-following control systems may lead to significant tracking deviations.Reinforcement learning,which enables interactive learning with the environment,has the potential to mitigate the impact of external disturbances on unmanned surface vehicle path-following systems.In this context,this thesis investigates reinforcement learning-based path-following control methods to enhance the tracking accuracy and adaptability of the wave glider in uncertain ocean environments.First,a comprehensive review of the research progress on the wave glider and their path-following control methods is conducted.The significance of reinforcement learning in wave glider path-following control is discussed in light of its application to USV motion control.Additionally,the wave glider dynamic model is developed,and a shore-based monitoring system is implemented to support subsequent simulations and sea trials.Second,a heading controller for the wave glider is designed.The proximal policy optimization algorithm is employed to dynamically adjust the parameters of the Proportional-Integral-Derivative controller,thereby improving its performance under varying conditions.Simulation results indicate that,compared to traditional optimization methods such as particle swarm optimization and backpropagation neural networks,the parameter optimization based on proximal policy optimization enables the controller to effectively maintain tracking accuracy while mitigating external disturbances.Third,an time-varying lookahead-distance strategy based on proximal policy optimization is proposed,building upon the conventional line-of-sight guidance method.This approach allows the agent to adjust the lookahead distance in real time based on tracking errors,thereby enhancing the dynamic performance of the guidance law and improving path-following accuracy.Simulation results demonstrate that this method effectively reduces cross-track errors under different path and environmental conditions,leading to improved tracking performance.Finally,sea trials are conducted to validate the proposed method in real-world ocean environments.Experimental results confirm that the reinforcement learning-based path-following control system effectively enhances tracking accuracy and improves the adaptability of the wave glider to complex ocean conditions.The findings provide valuable insights for the further development and application of wave glider control strategies.更多还原