基于神经动态优化与模型预测控制的欠驱动船舶精确路径跟踪

Precise path following of underactuated ship based on neurodynamic optimization and model predictive control

  • 摘要:
    目的 旨在解决传统模型预测控制方法采用在线滚动方式进行优化求解,造成欠驱动船舶路径跟踪预测控制器计算量大的问题。
    方法 将神经动态优化系统引入模型预测控制方法,提出一种具有实时性的欠驱动船舶路径跟踪预测控制器首先,针对船舶欠驱动特性,采用并改进视线制导策略:针对传统视线制导策略的运动学模型不确定性问题,基于滑模思想,提出鲁棒视线制导方法;更进一步,针对外界干扰影响下船舶易产生侧滑角问题,对侧滑角进行补偿,提出鲁棒自适应视线制导方法,提高系统对模型不确定性与外界干扰的鲁棒性。其次,针对欠驱动船舶输入饱和问题,通过模型预测控制方法将船舶路径跟踪问题转化为含有输入约束限制的二次优化问题。最后,针对模型预测控制方法采用在线滚动优化策略导致计算负担增加问题,基于投影递归神经网络,建立神经动态优化求解器,通过并行求解含有输入约束限制的二次优化问题,提高计算效率。
    结果 经过直线和曲线路径跟踪仿真,验证了本文所提出的具有实时性的欠驱动船舶路径跟踪预测控制器能够达到任意路径跟踪的目标。对比仿真实验结果也表明所提方法相较于Fmincon优化求解器(MATLAB内置求解器)计算效率提升约90倍,具有显著优势。
    结论 研究结果对于提升欠驱动船舶路径跟踪预测控制的实时性能具有一定的工程实用参考价值。

     

    Abstract:
    Objective The traditional model predictive control method employs a repeated online optimization approach, resulting in a high computational burden for underactuated ship path-following predictive controller. To address this issue, this paper presents an efficient predictive controller for underactuated ship path following based on the neurodynamic optimization system.
    Method  First, the line-of-sight (LOS) guidance principle is employed to mitigate the underactuated problem herein; for kinematic model uncertainty in traditional LOS guidance law, a robust LOS guidance method based on the sliding mode concept is proposed. Furthermore, the sideslip angle induced by external disturbances negatively affects path following. To compensate for this effect, a robust adaptive LOS guidance method is proposed, enhancing robustness against model uncertainty and external disturbances. Second, in order to address the input saturation problem, the model predictive control is adopted herein to transform ship path following problem into the quadratic optimization problem with input constraints. Finally, the neurodynamic optimization solver is proposed based on the projection recurrent neural network herein to solve the quadratic optimization problem with input constraints, enhancing the computational efficiency.
    Results In this study, both simulations for straight line path following and curved line path following are conducted. Overall, the simulation results show that the presented efficient predictive controller can achieve arbitrary path following. Additionally, the comparative simulations are performed, revealing that the presented method exhibits advantage in computational efficiency compared to the Fmincon optimization solver. Specifically, the neurodynamic optimization solver achieves approximately a 90-fold improvement in computational efficiency compared to the Fmincon optimization solver.
    Conclusion The research results have practical value for improving the real-time performance of underactuated ship path following. In the future, the proposed real-time predictive control method will be extended to the application of multi-ship cooperative predictive control.

     

/

返回文章
返回