Research On Universal Approximation Of NFE Model

As the three major intelligent technologies, Fuzzy Reasoning (FR), Expert System (ES) and Neural Network (NN) are integrated with each other to construct a variety of high-performance intelligent systems, and are widely used in various multi-source information fusion systems. Modern information fusion system is inseparable from these technologies. At present although there are dozens of fusion algorithms, the reserch of the basic framework of intelligent fusion algorithm is still scarece. Besides, there is no universal, transparent intelligent information fusion model currently. This article will focus on the universal approximation properties and generalization ability of a tightly-coupled model named NFE model combined with NN, FR and ES.Firstly,this article introduces the related theory and the current research situation of three major intelligent technologies. And the two dimensional models which are based on two of the three major technologies lack the capabilities of handling problems comprehensively. Meanwhile, it analyzes and compares the shortcomings of not representing the knowledge of two-dimensional model structure and the abstract symbols. And it proposes the necessity and feasibility of building a comprehensive three-dimensional model of intelligent information fusion according to the analysis.Secondly, a class of information fusion NFE model is studied. According to a detailed analysis of this model using Weierstrass theorems, we can know that this model can approximate any arbitrary accuracy for nonlinear systems, and it has the universal approximation. So it can be regarded as a class of complete intelligent universal approximator which can deal with uncertainty information fusion of complex environmental background well. In addition, the thesis analyses the good generalization ability of NFE model, and it make a simulation using this model to approximate nonlinear functions, compare the simulation results with other models.Finally, information fusion NFE model is applied to bistatic HFSWR (high frequency surface wave radar) track initiation in this paper, and it has built the fuzzy relationship between the radial velocity change, radar cross section change and signal to clutter ratio magnitude and the degree of belief for the target measurement which is regarded to confirm the update in the proposed method and the position data are utilized indirectly to choose the corresponding member functions of input variables as the judgement conditions of the expert system, so as to solve the low track initiation probability problems under the circumstance of intermittent measurements and low data transfer rate in bistatic HFSWR system. The research result indicates this method is applicable to the whole detection region.