A Study On Model-Driven Decision Support System

Model-driven decision support system (MD-DSS) is the most important type of decision support system. Nowadays, the model management and the development of MD-DSS has become the core research focused by most of researchers domestically and overseas. However, there are lots of shortcomings and limitations about application of MD-DSS in the existing outcomes. The model representation and management in MD-DSS cannot support decision making effectively, and there is not still a satisfactory and complete solution for rapidly developing MD-DSS, which makes applicability and developing efficiency of MD-DSS cannot meet the practical requirements of the firms. It also gives rise to the low popularization and application level in the firms.In this dissertation, MD-DSS is systematically studied on above issues. The major contents we have researched are included as following:(1) Study on model representation in MD-DSSBased on analyzing the basic characters of MD-DSS and the general process of solving decision problems, four principles are brought forward for design of model representation, such as supporting whole process of problem solving, uniform representation of model and algorithm, separating of model, algorithm and data, and user-oriented. According to these principles, a new method of problem solving based model representation (PSBMR) is put forward. From five layers, such as scheme layer, task layer, model layer, data layer and algorithm layer, the whole process of decision problem solving can be fully represented, variant model and algorithm classes can be represented uniformly, the management means of models are clearer and the capacity to reuse schemes, models and algorithms is higher than ever. (2) Study on model management in MD-DSSStudied some important problems of model management in MD-DSS based on PSBMR method, and also put forward the corresponding solutions. Based on the connection of model and data, it is easy to integrate the models to overcome the granularity confliction and cyclical confliction in model integrating. Based on metadata management, multiple databases can be connected with model conveniently by professional users. A set of expression rule is established, which enables model management system to verify and understand model expression. The class defining of algorithm information and algorithm parameter helps to standardize the interface between model and algorithm. Any model can call any algorithm in a uniform mode, and the algorithm base can be extended by any user and any third-party software manufacturer. By collaborating among three solving layers, MD-DSS can manage the process of decision problem solving.(3) Study on fast development of MD-DSSBased on PSBMR method, a general MD-DSS architecture consisting of four bases (scheme base, model base, algorithm base and database) and six managers (scheme base manager, scheme manager, model base manager, algorithm base manager, data connection manager and metadata manager) is designed. Then the object-oriented approach is introduced into the development of MD-DSS from the model representation, meanwhile, a complete solution of component-based development for MD-DSS is put forward. A component system, which consists of a series of components such as data connection component, metadata management component, scheme management component, model management component and algorithm management component, is designed for fast development of MD-DSS. By using PSBMR method and the MD-DSS development components designed and realized in this dissertation, an actual MD-DSS, the Decision Support System for the Metal Resources in China, is developed. The result shows that the component-based development not only increases the development efficiency, but also enhances the individuality and the flexibility of MD-DSS compared with other development methods.