A Study Of Resource Scheduling Based On Multi-Agent Coordination

Resource allocation is an important type of decision making activity in project planning. Actually, group decision in project planning has two attributes. Firstly, the plan is made by many specialists together. Secondly, the plan must be executable under constraints as resource limitation. It is urgently needed of theory and method to support the coordination in this kind of problem. The traditional approach generally models this problem with mathematic programming, where the central decision-maker has to collect all information, and the model can hardly be changed once it is constructed. It has a lot of limitation in practice. Especially it is not easy to be applied in the open environment where information is incomplete and inaccurate.The swift prevalence of internet has made coordination based on software agent being a new research hotspot. There have been some simple applications like schedule management. The demand for coordination comes from the distributation of resources, entities, information and the interdependencies among them. The characteristics of group decision on resource allocation just bring this kind of properties and interdependencies. Firstly, the knowledge and information needed in the decision are distributed among each decision-maker; therefore no one can see the global view of the decision problem easily. Secondly, the decision-makers'specialities and capabilities are various, so the percepted information are different and even inaccurate. Thirdly, the decision-makers'preference are unlike and their choices of objective are dependent on their own preference. Lastly, there exist privacies in reality, which make it hard to exchange all information. Thus, MAS coordination is more applicable to group decision. This dissertation studied the MAS coordination applied in project plan integration with resource constraint.In the first place, the research basis and background is introduced, which include theories of MAS coordination, current methods, technologies and applications. The theories of MAS coordination is set forth from the review of agent and MAS study, coordination model. Current theories, methods and technologies are introduces, and their properties are analysed. The several mainstream coordination models are also discussed. The content and significance are proposed as the end of introduction.Individual rationality is a foundation of social intelligence. Based on the analysis of agents'characteristics and nature, several mainstream infrastructures of agent are discussed, especially the BDI framework. Then a new method to construct individual rationality is presented. Compared with traditional approach, this method is more applicable to open complex system with its dynamical and robust properties.Modern AI is focused on social intelligence. The individual rationalities converge on higher and complicated intelligence during social interaction. Without right support of social interaction, even more complicated individuals can hardly complete complex task. Thus, this dissertation treats multi agent interaction and coordination game as keystone. From the simplest interaction to corperation, each level of interactions between agents is analysed. The communication in interaction is also introduced. Then, coordination method based on coordination knowledge level is discussed. The coordination model is analysed using coordination game.Design based on agent and MAS is easy to be understood and implemented, and has merits such as load-balance, fault-tolarence, and adapted to dynamical context, which are more applicable in practical project resource allocation. Based on forementioned individual rationalities and social interaction, this dissertation focused on resource constrained multi-plans integration problem, and exploit a new method of resource allocation. Firstly the context where multi-plans run simultenously with limited resources is analysed. Then coordination model and agent communication are described. Since individual decisions is affected by the users that agent representats, individual decision making process is emphasized. MAReCo is proposed based on these thought. With three reasonable assumptions, namely individual rationality, information exchange and MAS negotiation, a formal description and solving approach are presented.To the implementation and practice of MAReCo, this dissertation taking transit scheduling for example, set forth the application background, system architecture, design of coordination rules and coordination objects, and MATs is constructe. Through numerical test, the feasibility and effectiveness of MATs is verified.This study on resource allocation has significant value to supply chain management, intelligent transportation system, and intelligent software.