Research On Nonlinear Filtering Based On Fuzzy Neural Networks

A mass of the advanced filter's algorithms is developed on the foundation of all kinds of essential algorithms and the better effects can be gained by improving the basic algorithms in special conditions. However, so as to accommodate to the identities of the noise's nonlinear and complexity, therefore the mechanisms of self- adaptation, self- organization and self-study are introduced to the filter to form the filtering method based on FNN (Fuzzy Neural Network) in this paper.FNN is a cross subject, which can complete difficult intelligent analysis and process effectively nonlinear, fuzzy and uncertain problems with simple mathematic methods and logical reasoning on basis of the combination of FL (Fuzzy Logic) and NN (Neural Network).In this paper, based on the research and analysis of the theories and algorithms of the classical BPNN, RBFNN and FNN, the initializing method of FNN is presented based on excitation function transformation distinct region in order to make the network model quickly trace the observed system and converge the optimization solution to the problems. The configuration of FNN is also presented by combining FL and NN based on their equivalence. FZZ methods have been approved availably in the application of the practice data of the seismic exploration .On the other hand, aiming at the different noise with the different characters, these methods such as median filtering, BP filtering and RBF filtering are applied to assess filter's capabilities. The outcome of the experiments sufficiently proves the validity, robustness, feasibility and practicability of the nonlinear filtering methods based on FNN.