基于迭代学习和神经网络的船用起重机控制

Controller design of an offshore boom crane utilizing iterative learning and neural network

中文关键词  船用起重机  迭代学习控制  神经网络  未知干扰

英文关键词  offshore boom crane  iterative learning control  neural networks  unknown disturbance

中文摘要

      作为一种重要的海上作业装备, 船用起重机被广泛应用于海洋工程的各类场景中. 然而, 船用起重机是一 类复杂的非线性欠驱动系统, 存在摩擦、未建模动态等干扰, 为控制器设计带来了巨大挑战. 更糟糕的是, 船用起重 机还面临海浪、大风等未知干扰的影响, 使得实际控制更加困难. 如何稳定高效地控制该类系统, 目前仍处于初步 探索阶段. 为了解决上述问题, 本文提出了一种基于迭代学习和神经网络的控制方法. 具体来说, 首先将未知干扰 分为周期与非周期两部分. 对于周期干扰, 利用周期估计器解决了对未知周期的估计问题, 在此基础上通过迭代学 习对干扰进行补偿; 对于非周期干扰, 使用双层神经网络进行逼近和补偿, 并设计了权重的更新律; 在补偿未知干 扰后, 基于反馈线性化设计了控制输入. 通过Lyapunov分析方法, 可以证明期望平衡点是全局有界的. 最后, 在所搭 建的船吊实验平台上进行了大量实验, 充分验证了所设计控制方法的有效性与鲁棒性.

英文摘要

      Offshore boom cranes are complex nonlinear underactuated systems, which are widely used in various scenes in marine engineering as important offshore operation equipment. Suffering from frictions and unmodeled dynamics, the control of offshore boom cranes has always been a great challenge. What’s worse, offshore boom cranes are extremely vulnerable to unknown disturbances such as sea waves and strong winds, making the corresponding control problem even more difficult. It still remains an open problem to efficiently control such complex systems. To address the above problems, a controller based on iterative learning and neural network is proposed in this paper. Specifically, regarding periodic disturbances, a period estimator is used to estimate the unknown period, based on which iterative learning is used to approximate and compensate for the disturbances; As for non-periodic disturbances, a two-layer neural network is used to compensate them with carefully designed update rate of the weights. Based on the compensated dynamics, the control input is designed utilizing the feedback linearization technique. Through the Lyapunov-based analysis, it is rigorously proved that the desired equilibrium point is globally uniformly ultimately bounded (GUUB). Finally, sufficient experiments are implemented on the selfbuilt offshore boom crane testbed to convincingly verify the effectiveness and robustness of the proposed control method.