Time-Varying Parameter Modeling Based On Particle Filter And Its Applications

Nonstationary signal processing is an important branch in modern digital signal analysis. As an effective tool for dealing with nonstationary signal, adaptive time frequency analysis based on time-varying parameter modeling is rapidly developed and widely used. When the stationarity of the signal is pool or even model is nonlinearity, how-to complete the signal modeling is especially focused at present.In this thesis, the modeling methods applying particle filter is investigated based on the existed theory of time-varying parameter estimation. Firstly, to limit the particle degeneracy, a new method of Kalman particle filter is proposed by introducing forgetting factor. Second, to improve the performance of the parameter estimation, another improved particle filter is developed using the state vector extension in auxiliary unscented. The analysis of the numerical characteristics of the methods is completed by computer simulation. The results show that there are many advantages for the methods presented here comparing to the traditional approach.Focused on the features and requirements of flutter test data processing, the paper's method is introduced for flutter boundary prediction under turbulence excitation. Simulation and flutter test of low-speed wind-tunnel of one aeroelastical mode are used to examine the specifics of the scheme. Especially, the technique can be expanded to the flutter test with variable speed.The related software is developed on Matlab6.5 and Labview7.1.