Study Of Aircraft System Identification Based On Neural Networks

As one of main branches of the automatic control and signal processing, system identification has played a more and more important role in many fields, such as space-flight, aviation, engineering control, chemistry, finance and so on. Now the subject of modeling and identifying various nonlinear systems has attracted many researchers. However, it is difficult for the classical identification methods based on the linear system or intrinsically linear system theory to provide satisfactory results. A Neural Network (NN) is a model with high non-linearity, which is able to uniformly approximate any sophisticated nonlinear relationships in any accuracy. So in recent years, there has been much activity in how to apply NN to system identification. In this paper, theories about NN are reviewed, and aspects of system identification using NN are discussed. The computer simulation experiments show the power ability of NN in modeling linear and nonlinear systems.At first, theories about system identification are systematically reviewed, including its contents and process. At the same time, the classical system identification methods are summarized, the advantage and disadvantage of these methods are analyzed. Secondly, the history of NN is briefly reviewed, and the structure, function and algorithms of feedforward NN and the recurrent NN are particularly introduced. Further more, the matters involved designing NN in practical application is presented. A dynamic nonlinear system is identified through recurrent NN using Back propagation (BP) algorithm and dynamic BP algorithm respectively. Compared with the classical method, the identification accuracy and the generalization capability of NN are testified to be superior in either the free-noise or noisy case. In the end, an aircraft system is simulated by software MATLAB, and the data of 3 subsystems are obtained. Then the identification of these systems is carried using identification models with multi-NN. The result shows the models can well exhibit the characteristic of systems.