Model Normalization And Multi-agent Domain Organization Model

Taking model and Multi-agent techniques as research background,this thesis makes a focal research on the implementation and application technologies of the model normalization,the algorithms of model generation,and the organization model of Multi-agent systems. The primary contributions of this thesis include:With respect to the research on the model normalization,model coding normalization,model data normalization,and model encapsulation have been achieved. Model coding normalization conveniences model management and retrieval;model data normalization includes the model description file (MDF) and the operation on the MDF,which separates model description and model implementation,unifies model expression,normalizes model data and model operation,and so on;Model encapsulation supports models programmed by any kind of language running on the same operation systems,and conforms them to normalization;Model normalization provides the basis for efficient management,organization,operation and encapsulation of models used in DSS,which makes them easy to understand,use and combine. Model normalization has also been applied efficiently in the project "The Space Decision Support Systems Development Platform".Component technology is the development direction of model programming technologies. As one of the important parts of model normalization,model component gives support to the share and reuse of models. This thesis designs the interfaces of model component for Decision Support Systems,which include model normalization interfaces,model management (including model static and dynamic management) interfaces,component - based DSS model modules and so on. Creation,representation,information retrieval,and operation of model components have been implemented in practical application. This part of work has also been used in scientific research projects.Model generation is one of important branches of machine learning. Based on thorough study on FDD algorithms,this thesis presents the algorithm of Multi-Dimension Formula Discovery from Data (MD-FDD). By introduction of derivative rule,derivative formula can be found;Through defining multi-dimension function space and reorganizing the function rules,the inaccuracy rules,and the end rules,the scope of formula finding is broadened;By introduction of three-dimension expand rule and multidimensional expand rule,multi-dimension formulas can be discovered;Through improvement of flow of algorithm,more complete traverse has been achieved. Compared with the BACON and other traditional FDD systems,the quantity and quality of the found formulas are both improved significantly. In addition,the algorithm of formula discovery based on genetic programming (GP-FDD) hasalso been put forward. The efficacy has been proved by practical result data.A major development direction of DSS toward network is building Multi-agent System with agent technologies to improve the system's abilities of solving problems. Through research on the organization of Multi-agent Systems,the thesis presents the Multi-agent Domain Organization (MADO) model,provides the architecture of domain organization,and the framework of compositive agent and functional agent in the domain. Based on MADO,the communication mechanism and collaborative model among agents have been put forward. Based on the need of military intelligence analysis application,MADO model has been applied in the modeling and development of "Military Intelligence Analysis Decision Support Systems based on Agent Collaboration (AC-IADS)". This is the first time for our military to use agent technologies to develop general decision support systems in the battlefield situation analysis.The achievements of this thesis have great theoretic and realistic significance in expanding application domain of decision support techniques,promoting model normalization,implementing model share and reuse and facilitating the application of agent technologies.