CORC  > 自动化研究所  > 中国科学院自动化研究所  > 毕业生  > 博士学位论文
题名智能控制中的学习与适应性--理论、方法与应用研究
作者徐滇生
学位类别工学博士
答辩日期1993-06-01
授予单位中国科学院自动化研究所
授予地点中国科学院自动化研究所
导师疏松桂 ; 毛绪瑾
学位专业控制理论与控制工程
中文摘要智能控制理论被誉为第三代控制理论,是目前国际控制界和人工智能界着重研究的 课题之一。随着其理论和方法的不断完善,智能控制必将对广大的控制领域产生深刻的影 响。 本文主要研究智能控制中的学习与适应性及其应用问题。全文由四个部分组成。 第一部分为综述部分,回顾了控制理论的发展历程,讨论了现有控制理论的局限性和 智能控制理论出现的必然性。按模糊控制,分级递阶控制,专家控制和基于神经元网络的 控制对现有的智能控制方法进行了分类,阐述了各自的基本思想和处理方法,分析了它们 的不足之处,讨论了需进一步加以解决的问题。最后对作者的工作进行了简介。 第二部分为理论及方法部分,包括二、三、四章。 第二章对智能问题进行了讨论。从认知科学的思想出发,提出了智能系统中智能行为 的表现形式,即系统的适应能力、学习能力、记忆联想能力及结构进化能力等,其中结构进 化是智能行为的高级表现形式。由此提出智能控制系统应具有的基本功能,并利用一般系 统理论方法对这些功能进行了形式化描述。 第三章通过将控制过程和生物遗传过程相对应,首次提出控制系统的遗传结构假设, 为在更广义的框架下研究智能问题奠定了基础。在此假设下通过引入遗传算法并加以修 改,可对控制系统的建模、控制参数优化,控制系统的结构进化进行统一处理。以此为基础 设计了一个软件系统。提出一种准模型方式下的评价方法,可用于系统的在线优化/进化 过程。最后提出了并行直觉控制思想。 第四章研究了智能控制中的知识自动获取问题。首次将机器学习方法中的分类器系 统用于控制知识的获取中,这对构成具有优良性能的智能控制系统具有重大意义。主要考 虑了以监督学习方式获得控制器和被控对象性能以及以非监督学习方式自动获取控制知 识,并研究了控制知识的泛化和提取问题。通过对倒立摆系统控制的研究,验证了方法的 可行性。 第三部分为应用部分,包括五、六章。 第五章基于扩展修改后的黑板模型结构,并结合自控系统的具体要求,提出了一种智 能控制系统框架。该框架是常规控制系统和各类智能控制系统的有机统一,其显著特点是 功能扩展的灵活性。用基于iI己Mx实时操作系统的PL/M语言初步实现,具有一定应用 价值。 第六章研究了循环流化床锅炉的智能解耦控制。利用分类器系统构成补偿环节,无需 对象数学模型,能通过学习逐步改善其性能。主回路的控制则采用直觉并行控制方
英文摘要Intelligent control theory is known as the third generation control theory, and is one of the major research subjects in control and AI circle. With the advance of its theory and methods, intelligent control will have a deep influence upon vast industrial field. This dissertation mainly deals with learning and adaptation in intelligent control sys- tems. The whole thesis consists of four parts. In the first part, i.e. chapter 1, an introduction to the development process of control theory is given, and the limitation of available control theory and the necessity of intelligent control are discussed. By the order of fuzzy control ,hierarchical control,expert control, and neural network based control, intelligent control methods are classified, and the fundamen- tal thoughts and principles of each method are expounded, faults are analysed, problems available are discussed. At the end, the work of author is briefly introduced. The second part, including chapter 2"--4, deals with theory and corresponding method of intelligent systems, mainly about adaptation and learning. Chapter 2 begins with a discussion about intelligence. Starting from cognitive science train of thought, the manifestation of intelligent systems is presented, namely, the ability to adapt to new circumstance, to learn from past experience, to remember and associate one thing with others, and to evolve structurally. In all of these, the ability of evolution is be- lieved high level form of intelligence. The fundamental functions a intelligent control system should possess are explicated, and are formalized by general system theory. In chapter 3, by corresponding control process with organic evolution process, a genetic structure hypothesis of control systems is proposed for the first time, and this will lay a foundation for the settlement of intelligent problems in a broad sense. On this hypothesis, and through the introduction of genetic algorithm to control field, the problems of system modelling, parameter optimization, and structure evolution of control systems can be settled consistently. On the basis of above results, a software package is designed. For the purpose of on-line optimization/evolution, a quasi-model based performance evaluation method is proposed and proved effective. At the end of this chapter, the thought of parallel intuitive control is put forward. Chapter 4 deals with the problems of knowledge acquisition in intelligent control sys- tems. For the first time, classifier system in machine learning field is used for the acquisitionof control knowledge, and that will be of great significance to construct intelligent control systems with excellent performance. Two main problems, i.e. property imitation of con- trollers and plants by supervised learning and control knowledge acquisition by unsupervised learning, are considered. The methods of knowledge generalization and extraction are also stu
语种中文
其他标识符264
内容类型学位论文
源URL[http://ir.ia.ac.cn/handle/173211/5634]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
徐滇生. 智能控制中的学习与适应性--理论、方法与应用研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所. 1993.
个性服务
查看访问统计
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。


©版权所有 ©2017 CSpace - Powered by CSpace