Research And Application On Multi-Source Information Fusion Based On Evidence Theory

With the rapid development of traditional Internet and mobile Internet, information construction of C hina has gradually entered the 4C revolution stage. The information provided better and faster services for the production and living of people. Meanwhile, some problems such as the overload, multi-source, complexity and uncertainty of information hinder the process of information construction seriously, bringing new challenge for the study of information science and information technology. Multi-source information fusion is important ways for information processing, handling, synthesizing and application, having great application prospects in networking, robotics, intelligent control and other fields. For the ability to adapt to uncertainty situation, evidence theory, which is a representative method of multi-source information fusion, has become one of the hot researches in recent years.Rely on the related topics of The National 863 major projects” CatFishAction”, this dissertation make further research on the uncertain information fusio n method based on evidence theory. Aims at improving the reliability and flexibility in application of evidence theory, this article focuses on three aspects as the BPA generation method, conflict evidence combination and BPA probability transformation, and makes the necessary improvements on it. The content is as follows:1. In order to solve the problem that the fuzzy similarity used in BPA of evidence theory exist paradox when the cognitive objective is a single point, a method to determine similarity and the basic probability assignment(BPA) generation based on bell function(Bell-F) is proposed. This method can avoid the paradox above which is inconsistent with the actual result in common sense, simultaneously it do more granularly describe similarity evaluation model for the reason of improving the accuracy of cognitive description further. The recognition accuracy, which an experiment is performed for data set of Iris, is 93.3%. The BPA on low identification properties can provide more reliable value compared with other methods. Furthermore, this method still has good adaptability in uncertain environment when the number of known samples is insufficient.2. Many improved methods of traditional Dempster-Shafer theory focus on the evaluation and handling of evidence conflict, but ignore the actual physical meaning of the application of the scene, so there are some limitations and unable to form a consistent solution. Aiming at this problem, this paper takes re-discussion of conflicts and paradox in the view of specific scene, for the reason that to study rationality from the actual physical meanings of combination rule. Then builds the mapping between application scenarios and combination rule, and put forward a classification method of combination rules based on scenarios. Finally, a compatibility classification was presented. With the experiments and analyses of it, this method is proved to be effective.3. In the process of BPA probability transformation, the conventional method directly distributes assignment into the single target state by transformation formula, without making a distinction between the various states. The reality is the information quantity of different state levels are not the same, and distribution may not be limited to a changeless formula. So these cases should be treated differently. According to the analysis above, evidence quantity and transfer function are defined and a BPA probability transformation method based on hierarchical evidence quantity transmission was proposed, to solve the problem of the rationality and flexibility in existing method. After that, a reduction in the case of regular distribution and the direct transformation formula was given. The method is effective and flexible demonstrated by the experiments and comparative analysis.4. For the actual demand in the study of mimic security router(MSR), it is proposed that the implementation structure of multi- level fusion and closed- loop Cognitive-Decision Center(CDC) with a control center and supporting of backstage database. The algorithm design of CDC main modules was given in levels. It is from the question of router variants selection, and based on the study of the multi-source information fusion and evidence theory in this article. The test environment is set up to simulate test. The results show that design scheme is effectiveness and feasibility.