迭代学习控制在非严格重复时延系统中的应用研究

发布时间：2018-01-08 07:01

本文关键词：迭代学习控制在非严格重复时延系统中的应用研究　出处：《山东大学》2017年博士论文　论文类型：学位论文

【摘要】：本文针对迭代学习方法在非严格重复系统控制中的理论问题,开展迭代学习控制系统的鲁棒性、自适应性与收敛性分析研究,并佐以仿真与实验验证。诸多传统迭代学习控制方法要求被控对象满足严格可重复性,该前提是保证系统性能的关键因素之一。然而实际系统中存在的时延、不确定性和其他复杂因素的影响,以及非马尔科夫系统(特别是分数阶系统)本身记忆性的固有存在,使得系统难以满足严格可重复性,从而影响迭代学习控制效果,甚至保障系统严格可重复性本身的难度会远大于控制系统设计。因此,在实际问题中,当系统非严格可重复时,应用传统的迭代学习控制方法,其理论分析结果与实际应用效果差别较大。尽管可以通过适当引入反馈环节并利用鲁棒控制与自适应控制领域丰富的结论提高迭代学习控制系统整体的效果,但是如何从根本上消除被控对象的非严格重复性对迭代学习控制系统的影响具有十分重要的理论意义与应用价值,是对现有学习控制体系的有益补充。时延是导致系统不可重复的重要因素,因此时延系统本身即为一类典型的非严格重复系统。本文首先针对固定时延系统和时变时延系统进行研究,然后针对具有一般意义的非严格重复系统进行分析。本文研究并设计了几类具有代表性的迭代学习控制器:线性迭代学习控制器、非线性参数型迭代学习控制器和鲁棒自适应迭代学习控制器等,并将其应用于各类时延系统中。由于迭代学习控制作为一类典型的前馈控制,其鲁棒性能有待提高,因此本文将前馈控制与反馈控制相结合,设计鲁棒自适应迭代学习控制器,有效提高了非严格重复系统的控制效果。为提高迭代学习控制系统的自适应性,设计了参考值型学习律,在参考值轨迹不精确已知但满足一定规律的情况下仍可实现精确跟踪。本文的收敛性分析既包括收敛条件分析,又包括收敛速度分析。在收敛性分析中,虽然迭代学习控制的特点之一是不需要精确的数学模型,但是由于各类不确定性以及系统复杂性的增加,收敛性分析需要利用系统的某些结构信息、关键参数估计和控制器增益辨识,因此本文通过引入迭代学习辨识的方法,对某些关键要素进行有效分析,从而有助于设计性能更佳的控制器,进一步提高系统性能。特别是重点针对非严格重复系统(包括整数阶系统与分数阶系统)的预调进行研究,分析了动态系统在初始时刻之前的历史是如何影响当前控制系统的性能。同时,提出了预调与初始化的概念,从而在保证系统性能的前提下,通过无损预调提高迭代学习控制效果。综上所述,本文针对非严格重复系统的迭代学习控制领域亟待解决的若干关键问题进行了理论、仿真与实验研究。针对典型的非严格重复系统-时延系统以及具有一般意义的整数阶与分数阶非严格重复系统进行研究,分析收敛域以扩大收敛范围,通过引入反馈环节增强了系统的鲁棒性与自适应性,通过关键参数估计优化了迭代学习控制的收敛速度,通过初始化分析与无损预调进一步提高了非严格重复系统迭代学习控制的效果,搭建了理论与应用的新桥梁。本文通过大量仿真结果与机械臂半物理模型对本文结论进行仿真验证,通过NAO机器人实验与刀片磨削应用实验验证了迭代学习控制在非严格重复系统中的可行性,对迭代学习控制应用的推广具有较大作用。
[Abstract]:Based on the iterative learning method in non strict control of the repeated theory system, study the robustness of the control system of adaptive and iterative convergence, with simulation and experimental verification. Many traditional iterative learning control method can meet the strict requirements of controlled object repeatability, the premise is that one of the key factors the performance of the system. However, in the system delay, uncertainties and other complex factors, and non Markovian system (especially fractional system) itself and the memory of natural, can make the system difficult to meet the strict repeatability, thus affecting the iterative learning control effect, and even security system strictly repeatable itself the difficulty will be far greater than the design of control system. Therefore, in practice, when the system is non strictly repetitive, the application of traditional iterative learning control method, the The difference between the results of theoretical analysis and practical application effect greatly. Although it can be through the introduction of appropriate feedback and use of abundant robust control and adaptive control in the field of conclusion to improve the iterative learning control of the whole system, but how to fundamentally eliminate the non strictly repetitive object of iterative learning control system's influence has very important meaning and the application of the theory of value, it is useful for learning control systems. Time delay system is an important factor that can not be repeated, so the delay system itself is a kind of typical non strict re system. Aiming at the fixed time delay and time-varying delay system is studied, and according to the non repetitive systems with general significance carries on the analysis. This paper studies and designs several kinds of representative iterative learning controller: non linear iterative learning controller. The parameter type iterative learning controller and robust adaptive iterative learning controller is linear, and its application in all kinds of time delay system. Because of the iterative learning control as a kind of typical feedforward control, the robust performance needs to be improved, so this paper will combine feedforward control and feedback control, the design of robust adaptive iterative learning controller, improved control the effect of non repetitive systems. In order to improve the adaptive iterative learning control system design, the reference value of learning law, in the reference trajectory is not precisely known but certain rules under the condition can achieve accurate tracking. Convergence analysis in this paper includes both the convergence condition analysis, including the convergence speed in convergence analysis. In the analysis, although one of the characteristics of iterative learning control does not need precise mathematical model, but because of all kinds of uncertainty and complexity of the system The increase of the convergence analysis need to use some of the structure of information system, the key parameter estimation and controller gain identification, this paper by introducing a method of iterative learning identification, analyze some key elements, which is helpful to the design of controller performance better, further improve the system performance. Especially for non repetitive systems (including the integer order system and the fractional order system) of the pre research, analysis of the dynamic system in the initial time before history is how to affect the performance of the control system. At the same time, put forward the concept of presetting and initialization, and under the premise of ensuring the performance of the system, through non-destructive preconditioning iterative learning control to improve effect. In summary, some key problems based on iterative learning control non repetitive systems to be solved in the field of theory, simulation and experiment research For non repetitive systems. Typical time delay system and integer order and fractional order is of general significance of non repetitive systems, analysis of the convergence domain to expand the range of convergence, robustness and adaptability of the system is enhanced by introducing feedback, the optimal estimation of the iterative learning control convergence speed through the key parameters, the the initialization analysis and non-destructive pre adjusted to further improve the non repetitive systems iterative learning control effect, to build a new bridge of theory and application. In this paper, with the mechanical arm by a lot of simulation results of semi physical model to verify the conclusion of this paper, through experiment and application of NAO robot blade grinding experiments verify the feasibility of iterative learning control in non repeat the strict system, has great effect on the promotion of the application of iterative learning control.

【学位授予单位】：山东大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP13

【相似文献】