The Research Of Big Data Based Realtime Decision Support System

The arrival of the big data era puts forward new requirements to the development of decision support system (DSS). The situation had occurred that the basic theories and technologies of traditional DSS couldn't accustom the big data environment. DSS used by many industries had exposed serious problems in the era of big data with issues such as realtime decision, the architecture, data processing and storage. Now more and more researchers are beginning to study the combination of big data and DSS at the theoretical and technical level.This thesis researched big data based realtime decision support system (BDRDSS), on the basis of analysing the defects, new challenge and demands of traditional DSS in the big data environment. The primary work is as follows:1. The thesis studied the theories, models and architectures related with BDRDSS, which also analysesed the attributes of BDRDSS. Then based on these attributes, it gives the formal definition of BDRDSS, which can be used to illustrate the functions and features that DSS should have in the big data environment. These measures can lay the groundwork and indicate the direction for further study. Aiming at the shortcomings exposed in big data environment of traditional architectures of DSS, it designed the architecture of BDRDSS, which can work efficiently in a complex big data environment. And it also designed a general and layered service model of BDRDSS based on the architecture of BDRDSS.2. The two key technologies in BDRDSS that task scheduling algorithms and data stream load shedding strategies were studied. Firstly, it discussed the features of the system resource of BDRDSS and designed formal models for computing resource, computing tasks and decision tasks. Based on these, it designed DC-SS-FF algorithm and TC-SS-FF algorithm for decision tasks scheduling with the goal of optimal span. By means of experiment, it proved that both of them have good approximate ratio. Then, it classified the load shedding strategies for data stream in BDRDSS, combining with the characters that different decision tasks have different requirements on timeliness. It also proposed a dynamic load shedding strategy based on timeliness value of data and tested the executive effect of this strategy by experiment. Moreover, it discussed the influence caused by the key parameters on the timeliness and quality of decision-making.3. The thesis designed and developed a BDRDSS prototype, which takes financial domain as an instance. Based on the architectures, models and crucial technologies of BDRDSS studied by the thesis, it built a BDRDSS prototype used to quantitative investing. By realizing the instance system, it explored the specific combination of DSS and big data technologies at the practical level. The instance system, moreover, practically tests and verifies other research results introduced by this thesis.