Research And Applications Of Decision Support Technologies In Multi-Dimensional Data Environment

Model management is at the heart of Decision Support Systems (DSS). Its theory and application are not completely mature. So it is still in development. Data Warehouse (DW) and On-Line Analytical Processing (OLAP) provide a new way for DSS. The intelligent DSS integrated with DW and OLAP becomes a new type DSS. Since DW provided a novel multi-dimensional data environment for DSS, model management in multi-dimensional data environment becomes a new research subject in DSS domain. As a new powerful decision support tool, OLAP can analyze the multi-dimensional data in DW. Because of its dynamic and interactive, OLAP has to return the query results in the time users can accept. Therefore the response speed of query is the bottleneck of OLAP applications.The related technologies of decision support in multi-dimensional data environment are discussed deeply in the thesis. It includes model composition, the optimization problems of OLAP queries and the developing technologies of large real DSS etc. The main contributions of this thesis are as follows.1. Decision problems are solved by executing the corresponding decision models in DSS. For certain decision-making situations, a sequence of models are executed to answer a complicated problem. Composing a sequence of models (i.e., creating a composite model) in response to a decision problem is known as the model composition problem. Model composition is the core problem of model management in DSS. The thesis proposed a new approach of model composition in multi-dimensional data spaces. Based on a multi-dimension data model, the operations of multi-dimensional data set. such as Union, intersection and projection etc., is defined. The multi-dimension data (?)odel supports modeling of irregular dimensions by defining a partial mapping from child level to parent level. Multi-dimensional data interface is represented based on the data model. A 6-tuple with input interface and output interface, which are represented by multi-dimensional data interface, is defined. The domain knowledge is represented by the first-order logic. Furthermore, we presented an algorithm for model composition and execution and it can find relevant data set, select appropriate models, and automate model composition by searching the domain knowledge. The method has been applied to a practical project named Grain Management Information Intelligent DSS (GMI-IDSS). The running statistics of GMI-IDSS showed that the model composition approach can solve many problems efficiently and has good application prospect.2. Range queries and partial queries are two types of important queries in OLAP. This thesis summarized the related optimized technologies for range queries, which mainly include range sum and range max/min queries. To improve the response speed of multi-dimensional query and decision efficiency of OLAP, we discussed the partial queries, such as partial sum queries, partial max/min queries and partial top-N/bottom-N queries. The Rank Decision Tree (RD-Tree) is a data structure for processing partial max/min queries. A data cube need build and save a RD-Tree in advance so that the results of partial max/min query can be gotten quickly by searching the RD-Tree. In this thesis, two approaches are proposed to solve the optimization problems of partial top-N/bottom-N queries based on RD-Tree. One way is calling partial max/min queries algorithm for N times with the character that the efficiency will reduce when N increase. Another is the improved RD-Tree search algorithm. By modifying the searching conditions, the query results can be gotten in one searching procedure. Both of them are implemented and compared with the relative work. The improved RD-Tree search algorithm has the higher query response speed, because the change of N value hardly affects the query speed.3. The developing techniques of DSS are studied. Under the background of grain management, a DW-based intelligent DSS is designed and applied. The grain trade is the lifeline of the national economy. With the development of grain information management, mass of history data was accumulated. So we built a grain trade data warehouse and try to make use of these data to help management decision. Based on the grain DW, a special ETL tool is designed and implemented, the multi-dimension data analysis of OLAP and data mining are developed. Some main decision problems, for example, the decision of grain inventory update by turn, the decision of grain dispatching and transportation, grain prediction and so on, are extracted from the key procedure of grain management. And then we studied the related decision models, designed and realized the grain management model based system. Finally, GMI-IDSS is built, which can provides scientific and reasonable decision policies for grain production, purchase, sale, processing, storage and transportation etc. At present GMI-IDSS is successfully applied in more than 10 provinces or grain group corporations4. An important task of grain management is the dispatching and transportation, which is a vehicle routing problem. To solve the problem, we propose a two-stage optimization algorithm. In the first stage, all the feasible routes are generated by a graph-search algorithm. Because majority of the possible routes don't satisfy the constraints, the search tree is pruned according to those constraints. To minimize the route costs, we select the optimum routes from the feasible routes set. So an integer programming model is designed in the second stage. Since the feasible routes generated in the first stage are less, the amount of the variants and the constraints of the integer programming model are greatly decrease. Thus the model has higher running speed and more robust. The algorithm has solved the problem of grain dispatching and transportation effectively.Above all, this thesis combines the theory of DSS with the practice of DSS developing. The research results push the development of the decision support technologies in the multi-dimension data environment. The grain management information decision support system is a successful application of DSS in practice, which has an important reference value and significance about the applications of DSS in the future.