Research On Construction And Management Method Of Model Orienting Complex Decision-Making Problem

Problem solving is a hot issue in the research field of artificial intelligence and management decision. In the field of management decision, the research for theory of problem solving is mainly based on DSS. DSS supports decision makers to solve semi-structural and non-structural problems in the way of man-machine interaction by making use of models and data. It is such a computer system that supporting but not substituting decision makers to make judgments and heightening the effectivity but not efficiency of decision. Therefore, the purpose of DSS is to support decision makers to solve problems. However, problem solving must be with the aid of the models, which could be used to solve problems. Furthermore, the data requirement of problem solving is also determined by models. So model is the core of DSS. The running of DSS is driven by models. And the relational theories of model construction, representation and management are vital to the research of problem solving.With the development of science and technology and social economy, the decision-making problems become more and more complex. The features of nonlinearity, variable structure and variable parameter of complex decision-making problem lay higher claim to the research on the theories relating to model. The exist theories relating to model always appear insufficient while being used in solving increasingly complex decision-making problems. On the basis of these backgrounds, the theories relating to model construction, representation, management, etc. are discussed further and in depth on the basis of research fruits of people of the past. The specific content is as follows.(1) The general solving process of complex decision-making problem, which includes problem definition phase, modeling phase, problem solving phase and phase for explanation and evaluation of solution is given. The formal description for solving process of complex decision-making problem is given based on the general solving process of complex decision-making problem. All kinds of knowledge spaces involved in the solving process of complex decision-making problem and the maps between the knowledge spaces are described formally. The incompleteness of qualitative knowledge and quantitative knowledge when being used in solving complex decision-making problem is researched. Based on the incompleteness research, a multifaceted model system of complex decision_making problem solving is proposed. And the concepts for structure model, relation description model and solving model of complex decision-making problem in the multifaceted model system are illustrated preliminarily.(2) A meta-synthesis approach to modeling orienting complex decision-making problem solving is proposed based on the in-depth comprehension to the solving process complex decision-making problem and the basic thinking of meta-synthesis approach. In this approach, the modeling process is divided into three phases, which correspond to the construction process of structure model, relation description model and solving model of complex decision-making problem, based on the multifaceted model system of complex decision_making problem solving. At the same time, the domain knowledge and the problem knowledge, which are qualitative, and the model knowledge, the method knowledge, the tool knowledge and the operator knowledge, which are quantitative, are used synthetically to support these three modeling phases.(3) A hierarchical logic representation framework of models is given by drawing lessons from object-oriented thinking. This framework gives a complete representation approach to models in terms of model type, model instance, model component and model element. Model type defines a general pattern of model taxonomy and modeling capabilities. Model instance refers to individual problem-specific models. It is in correlate with the process of problem solving closely and is the instantiation of model type. Model component aims to decompose individual model instance into parts, which are small model units with certain basic functions, parameters and procedures. Each component comprises some model variables and their associated actions. Model element includes all the details of a specific model instance, such as variables, parameters and simple mathematical functions. They are basic objects used to construct model components. On the basis of the hierarchical logic framework, the construction approach of model component based on model element and the integration approach of model component are discussed. The algorithm for constructing model solving chain based on the depth-first search algorithm and the algorithm for model selection based on the satisficing strategy are put forward.(4) A hierarchical pattern for model storage and management based on relational database is put forward. This pattern is made up of basic layer, building block layer and conceptual layer. The basic layer describes the physical storage and management pattern of the model element, the model component, the model instance and the model type in the hierarchical framework of model representation. It can be used to support constructors of decision models. The building block layer encapsulates the details of basic decision models and supplies the basic information of decision models through interfaces of model building blocks. This enhances the reusability of decision models. The building block layer can be used to support the constructors of DSS. The conceptual layer establishes the connections between decision-making problems and model building blocks through giving the formal representation of decision-making problem. It can be used to support the decision makers and analyzers.(5) The taxonomy for agent in the problem solving method of multi-model cooperation based on Multi-Agent is given. Then, the approach to the construction of model agent, which includes direct development approach and the approach of using agent shell to encapsulate model software, is put forward. Additionally, the concrete problem solving flow of multi-model cooperation based on Multi-Agent is discussed.(6) A prototypical system for model management is researched and developed. The architecture as well as the function and implementation method of the components for the prototypical system are described. The concrete application of the prototypical system in the application of decision support technology of national economy and society development that of administrative area oriented, which is national key scientific and technological project of the tenth five-year, is discussed. Some models and components, including combinational time sequence model, combinational and multivariate emulation model based on neural network, spatial and distributed consumption requirement analysis model and spatial and distributed population simulation model, for decision analysis of regional national economy are designed and implemented. Then, the problem management for decision analysis and metadata management for component integration are discussed. Additionally, by means of an example that predicting the gross domestic product, the specific solving process of complex decision-making problem based on the model system is discussed. Furthermore, through the analysis of prediction result, the scientificity and practicability is demonstrated. At last, the application for decision support technology of pilot regions, which include decision technology application of west China development and pilot application of Chongqing city, is discussed.