Research On The Data Fusion Method And Its Application Based On RBFNN And D-S Theory

In recent years, with growing various application demand, data fusion technology as an emerging interdisciplinary has been developing rapidly and arousing wide concern. Large quantities of data from various sensors or sources of information are looked forward to complementary mutually going through data fusion system, to get and make more comprehensive, exacter and faster decisions and judgments on observed object.The work in this paper aims to further enrich and research the data fusion method in-depth, and combing the need og the subject, put the focus of the research in the application of network intrusion detection, specific contents and emphasis summarized as follows:Firstly we introduced the basic knowledge of data fusion related data fusion, introduced emphatically the hierarchical model and commonly used method of data fusion and analysed the advantages and disadvantages of them.Then we present the two kinds of method using universal in data fusion, that is neural network and D-S theory thoroughly, and focus the introduction on the basical principle, learning and training algorithm of the RBF neural network, also introduce the theory and synthetic rules related D-S evidence theory and study the nature with the synthetic rules. All of these is for the deeper reserch on these two methods and the application in intrusion detection later.Then in the fourth chapter of this paper, after introducing the data fusion method based on neural network, to solve the problem of existing redundant and useless information or fertures, we put forward the improvement method combing Fisher scores; And futher think up the two level fusion method combining the neural network and evidence theory, we also analyze the application with this method in network intrusion detection.In the fifth chapter, in allusion to the methods proposed in fourth chapter, we make some experiments in intrusion detection fields using them. Firstly some concepts involved in the experiments and the using data sets are introduced, then make relevant simulation, and analysis the results profoundly.The results in the last simulation experiments show that after discarding some low scores of some characters, Fisher_RBF proposed in this paper can not only ensure the detection rates, but also can make the rate of false positives get reduced; the two level fusion method of RBF_DS has better detection perfomance than that using neural network lonely.