Tham khảo tài liệu 'an introduction to intelligent and autonomous control-chapter 11: learning control: methods, needs and architectures', công nghệ thông tin, quản trị web phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | 11 LEARNING CONTROL Methods Needs and Architectures Mieczyslaw M. Kokar Department of Industrial Engineering and Information Systems Northeastern University 360 Huntington Avenue Boston MA 02115 kokar@ Abstract The first part of this chapter shows how architectures of control systems have gradually evolved from adaptive control to learning control. This evolution is explained in terms of the need to fulfill the requirements of greater autonomy and generality of control systems. Learning controllers should be able to learn three kinds of knowledge goals models and control laws. These three kinds of knowledge constitute the domain of intelligent control which uses methods from control operations research and artificial intelligence Al . The second part of this chapter is devoted to the COPER IC architecture being developed by the author. This includes an overview of author s work on automatic discovery of physical variables relevant to a particular model the use of physical similarity in the process of modeling integration of qualitative and quantitative techniques and the use of reinforcement learning for control. The chapter closes with an author s view of future needs and developments in the area of learning control. 1 INTRODUCTION This chapter is devoted to learning control. Our main goal is to show that the emergence of this approach to control has resulted from the need to satisfy a specific set of performance criteria. We focus our attention on two of such criteria autonomy and generality. Without being too formal we interpret autonomy as the ability to choose goals and devise new control laws. This interpretation is consistent with the dictionary 264 INTELLIGENT AND AUTONOMOUS CONTROL definition of autonomy as the right of self-government. Generality is understood as the quality of being applicable to the whole and thus a system is more general than another if it is applicable to more situations or domains. Antsaklis and Passino 1 discuss the
