【作者】 林坚；

【导师】 刘山；

【作者基本信息】 浙江大学 ， 模式识别与智能系统， 2007， 硕士

【摘要】 迭代学习控制作为现代控制理论的重要分支之一，其高精度的轨迹跟踪能力一直为人所称道。但实际控制系统中普遍存在的各种非标准条件，如含结构不确定性、输入输出扰动、初态偏移和轨迹变化等，往往使得标准条件下收敛的学习律无法满足原有条件，甚至于发散。这同时也影响到迭代学习控制的大规模工业应用。因此，研究在非标准条件下依然收敛的迭代学习律，并保证鲁棒跟踪性能，就显得十分重要了。本文着重研究在各种非标准条件下迭代学习控制的设计问题。针对含未知有界扰动的系统，采用前馈—反馈分部设计法，将迭代学习控制的设计问题转化为标准H∞问题后求解；针对系统存在结构不确定性问题，将前馈—反馈作一体化设计，并将该问题转化为鲁棒优化问题后求解；针对输入输出扰动、初态偏移等非标准问题，设计抗干扰的改进型算法；针对轨迹变化等非标准问题，采用轨迹学习的策略以获得最优输入初值的估计，从而加快收敛速度、改进瞬态响应性能等；最后以工业机械臂为仿真对象来验证算法的有效性。主要成果有：(1)考虑一类受未知有界干扰影响的线性系统，通过鲁棒迭代学习控制分部设计法来改善其跟踪性能；首先将该设计结构下的迭代学习律收敛条件转化为模型匹配问题，然后利用标准H∞方法求解；最后对线性化后的工业机械臂作仿真，验证了算法的有效性；(2)考虑系统存在结构不确定性问题，运用鲁棒迭代学习控制的综合设计法，将反馈控制器与前馈控制器作关联设计，并将其转化为二自由度的标准设计问题；而后分析该设计下的学习律收敛条件，并利用结构奇异值及μ综合等鲁棒控制相关理论求解；再考虑系统存在输入输出扰动和初态偏移等非标准条件问题，对原有算法作必要改进，并分析其收敛条件；最后利用针对直流电机伺服系统所作的仿真验证了算法的有效性；(3)考虑系统期望轨迹变化的问题，增加对“历史轨迹”的学习能力，从而达到改善系统跟踪性能的目的；具体采用了局部权重思想来逼近系统逆，从而求得最优输入初值的估计；最后通过对直流电机伺服系统的所作的仿真验证了算法的有效性。更多还原

【Abstract】 Iterative learning control (ILC) has been widely recognized by its unique capability in improving the robust performances of a control system, which usually based on the notion of repetition and learning. But good tracking performance employing conventional ILC strategy has largely relied on the assumptions and initial conditions, such as no structure uncertainties and disturbances, same or similar desired trajectories and so on, which largely limited its industry implementations. So how to apply the ILC strategy under nonstandard conditions concerns the present and furore of ILC.In this dissertation a robust iterative learning control strategy is proposed for against the disturbances, perturbation, noises and so on. For the issue of unknown but limited disturbances, one separated design scheme is presented for the open-and-close-loop structure. Then it is resolved as a standard H∞problem. For structured uncertainties and disturbances related issues, a synthesis method is utilized to choose systematically the parameters of learning and feedback controllers while achieving the optimization between robustness and tracking performance. In the presence of changeable tracking trajectories, a locally weighted method is employed for restructuring plant inversion which helps to estimate the optimized initial input.The main achievements are listed as followed.The first part mainly concerned with robust performance issues while a linear system is perturbed by unknown disturbances. By adding the ILC open-loop along to the original control system, it can improve the tracking accuracy while not destroying original system robustness. The design is then resolved as a comparable mold-match problem and a standard H∞method is employed for parameters design. The effectiveness of the method is demonstrated by the simulation of rigid manipulators.In the following part a two-degree-of-freedom architecture is employed for dealing with the issues of unstructured uncertainties, disturbances and so on. This synthesis method can systematically design the whole open-and-close-loop simultaneously which in return achieve the optimization between the robust stabilities and performances. The method of structured singular value andμsynthesis is employed for deriving the final parameters. A simulation on servo motor control system is conducted last for demonstration.The last part focused on the issues of changeable tracking trajectories. The importance of the selection of initial control input in error convergence is highlighted. The locally weighted theory is introduced for restructuring the linearized plant inversion while estimating the optimized initial input. This method is so general that it can be applied to most of ILC algorithms, including nonlinear ones. The computer simulation is also conducted on the servo motor control system for discussion.更多还原