Research Of Coordination Knowledge And Coordination Strategy Based On Multi-Agent Systems

Multi-Agent Systems (MAS) is an useful abstract model and research model to cope with complex natural systems and artificial systems with characteristics such as distributed, large, dynamic, open, intelligent, social, etc. Coordination is a very important part in MAS research. Only through coordination, agents can form large social structure and can show a certain content of sociality. Good coordination can improve efficiency, stability and practicability of MAS. At present, the research of coordination still remain several problems, such as the edge of coordination knowledge, an uniform research frame of coordination theory, the sharing and reuse of coordination knowledge and so on. These problems embarrass the use of coordination knowledge and the improving of coordination theory.Besides the complexity of coordination itself, the main cause of these problems rests with the close relationship between coordination knowledge and application environment. So, this thesis studies coordination knowledge in MAS domain and aims to construct a whole research frame of coordination knowledge in Multi-Agent Systems.To decide the edge of coordination knowledge is the precondition of MAS coordination research. At the beginning of this thesis, we analyze the necessary of studying coordination knowledge and use knowledge level methods to separate coordination knowledge and reality knowledge. And then, we present that coordination knowledge in MAS can be divided into action coordination and structure coordination. In order to grasp the essence coordination in MAS, we design a kind of knowledge ontology, called coordination ontology.Based on coordination ontology, we discuss the action coordination and its formalization. The management of interdependence is the key point of action coordination. In this part, we investigate the conception of interdependence, and divide interdependence into two kinds, interdependence based on time, interdependence based on resource. We use coordination ontology to formalize several basic interdependences and use the business flow of internet bookshop to explain the usage of coordination ontology in interdependences formalization.Then, this thesis discusses the action coordination strategy based on Petri nets and XML. In this part, we first study the conception and classification of action coordination strategy and use Petri nets as the formalization tools to represent it. And then, we use extended XML to translate Petri nets to computer readable language. In the end of this part, we design and realization this kind of translation software tool.After that, we study the structure coordination based on MAS. The management of interaction is the key point of structure coordination. In this part, we discuss the conception and main content of structure coordination first. Then, we use the case of internet bookshop to study three kinds of MAS structures, hiberarchy structure, flat structure and adhocracy structure, and discuss the influence of these structures on interaction in MAS.On the base of the above, we use Problems Solving Methods (PSMs) to study structure coordination strategies. We discuss three main structure coordination strategies, direct supervision strategy, standardization of work strategy and mutual adjustment strategy. Then, we use PSMs to formalize these strategies and through an experiment to compare the efficiency of these strategies.At last, based on the case of internet bookshop, we discuss the design of Coordination Reasoning Machine using coordination knowledge and implement the several functions of action coordination.