Font Size: a A A

Research On Multi-Agent Technology And Applications

Posted on:2003-11-22Degree:DoctorType:Dissertation
Country:ChinaCandidate:S H LanFull Text:PDF
GTID:1118360095452309Subject:Computer applications
Abstract/Summary:PDF Full Text Request
With the development of computer science, agent is playing a more and more important role in artificial intelligence and computer science. Researchers from both industry and academia are paying more and more attention to theories and applications of the agent.An agent with human being properties such as autonomy, sociality, adaptability and intelligence can act as a human. The application of agent systems covers every walk of life. With the development of information infrastructure, people have more requirements for high quality application. High intelligentized, high reliable, quickly adaptive and networked systems are the goals of the application. According with agent's properties these goals stimulate research work in agent theory and application.This dissertation aims at a bridge between agent theory and application. The main research works are as follows:A hierarchical multi-view agent system model is proposed in this dissertation. Taking reuse and easy-implementation into account, we divide an agent into different modules according to specific-general model, knowledge-process model and individual-social-world model. Then we specify and descript a multi-agent system model by four models: organization model, function model, information model, and communication model.The dissertation researches into user interface agent in multi-agent system. A framework of user interface agent is presented and some issues in implementation are discussed in the dissertation. Also several key mental status-based action rules are presented in the dissertation. In order to make the user interface agent adjust to evolvement of the system, the dissertation also proposes a collaborator selection method-multi-level top-n random selection algorithm that ensures the higher adaptability and flexibility.We study the negotiation in multi-agent system. Our research work deals with two different methods of forming cooperative relation. One is the famousnegotiation protocol-contract net protocol. We extend the contract net and introduce several parameters for mental state such as trust degree, conservativeness degree, cooperation frequency and activeness degree into process of announcing and bidding. We combine the BDI model of an agent with the extended contract net and implement contract net through a series of BDI-based rules of action. Another method to set up a cooperative relation is to form a coalition of agent. A coalition of agents is a set of equal and cooperative agents that take the task together. The method we propose take into account the historical records in coalition forming. Based on the set of successful cooperators, the method assures the better quality of completing task at a less communication and calculation costs and improve the efficiency of negotiation.As an intelligent entity with sociality, an agent should have capability of communication. Not only should an agent communication language work as a general communication protocol, but also needs to convey an agent's mental state. KQML is a popular extensive communication language for agents. In order to meet the needs of application, we extend KQML's performatives. Also we present an implementation of KQML.Based on the research mentioned above, we present a prototype of Agent-Based Distributed Expert System for Firearms Design. Through this prototype some of our ideas and design are validated.
Keywords/Search Tags:multi-agent system, BDI agent, interface Agent, Agent negotiation, contract net, coalition, agent communication language, expert system
PDF Full Text Request
Related items