An efficient model predictive control design for ship autopilot, which is a representative marine application, is proposed based on projection neural network in this article. Ship motion control at sea exhibits the characteristics of large inertia, strong nonlinearity, and large delay; furthermore, it is frequently influenced by the external disturbances, leading to a complex uncertain problem. In addition, the amplitude of control input—the rudder is constrained. Given the mechanism of on-line computing and the advantages of handling constraints, the model predictive control is one of the most favorable solutions for this problem. Nevertheless, the major challenge of the implementation of traditional model predictive control in application is the computation intensity. In this article, the capability of parallel computation of projection neural network is employed to optimize the objective function formulated by traditional model predictive control method, aiming to improve the computational efficiency. The overall information of ship motion is normally difficult to be obtained; therefore, a state observer should be also included. Extensive studies are conducted to illustrate the effectiveness of the proposed control design.

中文翻译：

---

使用基于观测器的模型预测控制的高效船舶自动驾驶仪设计

本文基于投影神经网络提出了一种具有代表性的船舶自动驾驶仪模型预测控制设计。海上船舶运动控制具有惯性大、非线性强、延迟大的特点；此外，它经常受到外部干扰的影响，导致复杂的不确定问题。此外，控制输入——方向舵的幅度受到约束。鉴于在线计算的机制和处理约束的优势，模型预测控制是解决这个问题的最有利的方法之一。然而，在应用中实现传统模型预测控制的主要挑战是计算强度。在本文中，利用投影神经网络的并行计算能力，对传统模型预测控制方法制定的目标函数进行优化，提高计算效率。船舶运动的整体信息通常难以获得；因此，还应包括一名国家观察员。进行了广泛的研究来说明所提出的控制设计的有效性。