Agent-based Control + Hybrid Systems + Realtime Parallel Systems

[Mecha]

Agent-based Control + Hybrid Systems + Realtime Parallel Systems

Đây là 3 chủ để mà Mecha đang đi theo. Trong quá trình ngâm cứu về nó, Mecha sẽ đưa lên đây những tài liệu và thông tin liên quan. Nếu anh em nào quan tâm thì nhào vô;)

[Mecha]

Đây là tóm tắt luận văn PhD của Albert J.N. van Breemen
"Agent-Based Multi-Controller Systems - A design framework for complex control problems"
Đây là một trong những nghiên cứu đầu tiên về Agent-based control.
Download luận văn này theo địa chỉ này:
http://www.ce.utwente.nl/rtweb/publications/2001/pdf-files/AJN%20van%20Breemen.pdf

Tóm tắt:
Practical control systems generally are systems that consist of multiple control algorithms; so called multi-controllers. Each control algorithm is designed to fulfill a particular task. Designing such systems requires engineering skills not discussed by control theory. These skills are structuring the control problem and organizing the individual solutions. This thesis deals with design aspects related to these two skills and presents theory and tools to support them.

In order to investigate the relevant aspects of structuring control problems and organizing individual solutions, a model of complex control problems has been formulated. This model describes complex control problems in terms of partial control problems and their interdependencies. It will appear that structuring control problems requires the identification of a “part-of” hierarchy of partial control problems, as well as the identification of the type of coupling between the partial control problems. The particular type of coupling between partial control problems determines how to integrate the obtained individual solutions.

To integrate multiple control algorithms, insight from the field of control engineering is needed. The Multiple Model Approach (Johansen and Murray-Smith, 1997) is a general design method for multi-controllers that consists of local controllers; i.e. control algorithms that are operational only in some limited part of the overall operating regime of the plant. While designing a multi-controller, the designer has to deal with particular integration aspects, such as deciding when to (in)activate a local controller, initializing and finalizing state variables of the local controller and combining control actions of several local controllers. The Multiple Model Approach relies on a supervisor to integrate the local controllers and deals with the integration aspects in a central way. In contrast with this, in this thesis a decentralized integration method is developed, based on the concepts of an agent and a multi-agent system. The result is an open design environment for multi controller systems, such that individual local controllers can be added, modified or removed from the overall multi-controller without redesigning the remaining system.

Combining the concepts of a local controller and an agent has resulted in a so called controller-agent. A controller-agent is a local controller that is responsible for the initialization and finalization of its state variables, has knowledge about its operating regime and has an interface to coordinate its behavior with other controller-agents. Controller-agents can be combined into groups by using coordination objects, which solves dependencies between controller-agents. As a result, the multi-controller system is structured as a “part-of” hierarchy, which in general reflects the decomposition of the complex control problem.

An implementation framework has been developed to realize an agent-based multi-controller. This framework is called the Multi-Agent Controller Implementation Framework (MACIF). This framework contains six functionally different software components, which can be connected to each other. These components can be specified by using the specification language MACSL (Multi-Agent Controller Specification Language), which was developed during this research.

[tuanphan]

Em đọc được khoảng hơn 70 trang, thấy có vẻ như phương pháp này là tổng quát hóa các phương pháp khác. Thấy có ví dụ làm nhiều subcontrollers và 1 bộ để quyết định lúc nào dùng subcontroller nào, hay bộ điều khiển phi tuyến phân làm 2 thành phần: 1 phần bù phi tuyến, 1 phần điều khiển hệ tuyến tính còn lại... Không hiểu nó có gì đặc biệt hơn (hình như tính tự trị và thông minh của từng agent), anh viết qua đoạn giới thiệu ngăn ngắn có được không?