On Optimum Nonlinear Filtering Via ANNs

This thesis explores the neural networks approach to optimum nonlinear feltering. The main research work is as follows:It analysis the actualities and the developing trends about the neural networks approach to optimum nonlinear filtering, itneural networks and the classical theory about the optimum nonlinearfiltering.Among them it is included not only the nonlinear Least Mean Square (LMS) estimation (spreaded Kalman filter), but the nonlinear Recursive Least-Square (RLS) estimation also.At the same time it analysis the feasibility of integration of neural networks and optimum nonlinear filtering.It discusses the Back Propagation (BP) network and the Radial Basis Function (RBF) network to optimum nonlinear filtering. With the MATLAB as software tools,simulations are made which applies the BP network and the Radial Basis Function (RBF) network to realism examples. A series of simulation waves are obtained, thus,the accuracy of the simulation model is validated by experiments.And the simulation verifies the superiority of the RBFs in nonlinear filtering such as shorter training time, least note and higher accuracy.Finally,the optimum nonlinear filtering based on Functional Link Artificial Neural Netwok (FLANN) is described,with its experimental results.It can be seen,the ANN used in the optimum nonlinear filtering may be a tendency in filtering of noise.