【作者】 逄勃；

【导师】 邵诚；

【作者基本信息】 大连理工大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 迭代学习控制从最初解决机器手运动的控制问题发展为解决很多具有周期特性的工程对象的控制问题,其利用实际输出和期望输出的偏差对当前的控制输入信号进行学习修正的思想,越来越得到了控制工程领域的广泛接受和高度重视。由于迭代学习控制方法所具有的不过分依赖模型的特点和实际应用中的良好效果,已经成为了现代智能控制技术中的一种重要控制方法。为解决传统参数优化迭代学习控制算法只适用于正定模型的局限性,本文提出一种高阶参数优化迭代学习控制算法。该算法利用多次迭代误差的信息,建立参数优化目标函数得到最优迭代学习律。在被控对象不满足正定性的条件下,仍保证跟踪误差单调收敛于零。此外,还提出了基于奇异值分解的PID型参数优化迭代学习控制算法,根据范数性能指标建立目标函数,通过对系统模型矩阵的奇异值分解得到学习增益矩阵,使算法应用于模型为非正定的系统时仍然保证闭环跟踪误差单调收敛至零。该算法还将PID型控制器引入到参数优化迭代学习控制算法的设计中,提高了学习速度。针对非线性系统跟踪控制问题,本文将拟Broyden法和参数优化迭代学习控制方法结合,提出一种新的优化迭代学习控制算法,利用拟Broyden法对系统的Jacobian矩阵进行迭代近似计算,并通过目标函数对学习增益进行优化。该算法不仅能够简化传统牛顿法中求逆计算所带来的复杂性,而且具有单调递减的特性和全局收敛性。针对一类特殊采样点的轨迹终端跟踪控制问题,本文提出了一种改进的牛顿法迭代学习控制算法,该算法具有单调收敛的性能和较大的收敛范围。利用一个间歇反应的生成物产量跟踪实例对算法的有效性进行了验证。本文将迭代学习优化控制方案应用到一类具有扰动和初始状态误差等不确定性的非线性离散系统中。从理论上证明了对于无扰动情况,算法能够保证系统跟踪误差以几何速度一致收敛于零,系统控制输入收敛到期望输入轨迹；对于具有扰动和初始状态误差的非线性离散系统,给出了该算法具有BIBO鲁棒稳定性的充分条件。本文还给出了P型迭代学习控制在带有扰动的一般性非线性系统具有鲁棒性的条件。从理论上证明了在初始状态误差和状态、输出扰动有界的情况下,系统输出能够收敛于期望轨迹的邻域内；在各种干扰消除时系统输出能够一致收敛于期望轨迹。更多还原

【Abstract】 Iterative learning control (ILC), which is firstly proposed in the movement control of manipulator, is a technique for improving the tracking performance of systems or processes that operate repetitively over a fixed time. ILC utilizes the deviation between actual output and desired output to modify current control input signal by learning, this control method has brought ILC widely acceptance and highly attention in the field of control engineering. The ILC algorithm does not rely on the repeatable system model very much and has good tracking accuracy in practice. Due to these characteristics, ILC has become an important control method of modern intelligent control technology.Traditional parameter iterative learning control algorithm ensures the tracking error converge monotonically to zero only when the original plant is positive. In order to solving this limitation, a high-order parameter optimal iterative learning control algorithm is introduced in this paper. The proposed algorithm is established by a quadratic performance index with the tracking errors from earlier trials to construct optimal learning law. The proposed algorithm can guarantee the tracking error converge monotonically to zero even the relaxation system is non-positive.A PID parameter optimal iterative learning control algorithm based on singular value decomposition is also given in this paper. The proposed algorithm establishes the norm performance index and obtains learning gain matrix by applying singular value decomposition to the original plant, and ensure the closed-loop tracking errors of this algorithm converge monotonously to zero even the original plant is non-positive. Furthermore, a PID controller is added to the design of ILC algorithm to improve learning efficiency.Focuses on the tracking problems of dynamic nonlinear system, a novel parameter optimal Broyden-like method based iterative learning control scheme is proposed in this paper. The Broyden-like method is used to iteratively calculate the approximation of the Jacobian matrix, and the parameter-optimal method is used to optimize the learning gain of the algorithm. Compared with traditional Newton-method, the proposed algorithm can avoid the calculation of the inverse of the Jacobian matrix of system. Furthermore, the algorithm has the properties of global and monotonic convergence.A class of specific target tracking problems for non-linear systems with the reference trajectory sampled only in some specific times is discussed. Accordingly, a modified Newton-method-based iterative learning control scheme is suggested to solve this problem. It is proved theoretically that the algorithm can ensure the closed loop system to have global and monotonic convergence. A case study of final product tracking problem of a mixed, liquid-phase batch reactor is given to demonstrate the effectiveness and tracking accuracy of the proposed algorithm.In order to solve robust iterative learning control problem of a class of discrete-time nonlinear systems with initial state errors, external disturbances and output noises, a novel norm-optimal based robust iterative learning control algorithm is proposed in this paper. It is proven theoretically that in the absence of these disturbances, the algorithm can guarantee the tracking error of the closed-loop system uniformly converging to zero geometrically. In the presence of these disturbances and uncertainties, the sufficient condition of robust BIBO stability for the proposed algorithm is given.Finally, this paper discus the robust stability condition of P-type iterative learning control algorithm for a class of nonlinear dynamic discrete system. It is proved that in the presence of state, output disturbances and initial uncertainties, the system output can converge to the neighbor domain of desired trajectories. In the absence of these disturbances and uncertainties, the system output can converge to the desired trajectories uniformly.更多还原