节点文献
不确定组合系统的若干控制问题研究
Study on Some Control Problems for Uncertain Composite Systems
【作者】 刘恩东;
【导师】 张嗣瀛;
【作者基本信息】 东北大学 , 控制理论与控制工程, 2005, 博士
【摘要】 现代社会日益面临着研究并处理规模庞大、结构复杂的系统问题(如工业过程、经济管理、生态环境、电网控制、交通运输、航空航天等),使得大系统理论已成为控制理论的一个专门领域。但是由于问题的复杂性,大系统理论的发展还没有形成一套完整的理论和方法。因此,设法利用系统自身的结构性质,对具有特殊结构的大系统采用特殊的研究方法,进而研究一般大系统理论不失为研究大系统理论的一条有效途径。在大系统理论中,由于组合系统自身的特殊性(由若干个子系统相互关联组合而成)和不确定性,确定了组合系统理论研究具有特殊重要的意义。 本文利用自适应神经网络控制技术,研究了非线性组合系统的跟踪控制问题和鲁棒镇定问题。即利用神经网络辨识非线性系统,设计相应的控制律和自适应律,保证闭环系统的信号是最终一致有界的,同时自适应律可以在线调节。另外,本文还研究了时滞不确定组合系统的鲁棒输出反馈控制问题,所设计的控制器对系统的不确定性具有很强的鲁棒性。本文的主要工作概括如下: 1.研究了一类非线性组合系统和不确定组合系统的跟踪控制问题,首先由动态神经网络分别辨识非线性组合系统和不确定组合系统,也就是利用动态神经网络逼近系统的未知项和互联项,其次设计控制器使实际系统的状态跟踪参考模型的轨迹。 2.讨论了一类非线性组合系统的观测器设计问题。组合系统可以通过动态神经网络建模,在观测器设计中,充分考虑了非线性组合系统中的互联项和动态神经网络逼近误差项对观测器性能的影响,增加了鲁棒控制项,这样所设计的控制器具有鲁棒性。 3.研究了一类组合系统的跟踪控制及性能分析的问题。利用高阶动态神经网络来逼近未知的非线性组合系统,设计了相应的控制器和自适应律;利用一些指标测试了控制器的性能。同时,克服了现存神经网络文献中一个主要局限性,即要求预先知道模型误差和最优连接权值的上界。
【Abstract】 Modern society faces studying and dealing with the system problems with large dimension and complicated structure (such as manufacturing process, economy management, environment, electrical net control, traffic transport, aviation ect), which make the theory of large scale systems become a special field. However, because of the complication of problems, there is no series of intact theory and method. Therefore, it may be an effective way to make full use of the structure of systems, and adopt special method for systems with special structure, then study the general large-scale systems. In the theory of the large scale systems, it is important to study the theory of the composite systems because of its special structure (composed of a series of interconnected subsystems) and uncertainty.In this dissertation, by means of the techniques of adaptive neural networks control, the tracking problem and robust stabilization of a class of nonlinear composite systems are studied. Nonlinear composite system is identified using neural networks, the control law and adaptive law are designed to guarantee the uniform ultimate boundedness property for the states of closed systems, and the weights of neural networks are updating on-line. In additon, the problem of robust decentralized stabilization is discussed for a class of uncertain delay composite systems, and the designed controller has strong robustness for uncertainty.The main work is as follows:1. Adaptive neural networks controller is designed to realize the tracking control for a class of nonlinear composite systems and uncertain composite systems. Firstly, nonlinear composite systems and uncertain composite systems are identified using dynamical neural networks, secondly the controller and updating law are designed to guarantee the uniform ultimate boundedness property for the tracking error and other