Particle Filter Based On Structural System Identification

With the health monitor of building structures being received increasing attention in recent years, the technology of structural parameter identification has become the hot researches at home and abroad., research in this paper aims to bring the sequential monte carlo (particle filter) method, which has a wide range of application in signal processing, statistics, and econometrics etc, into the structural parameter identification. The time varying systems can be stated in the form of a dynamic state space model. For linear models and Gaussian noise, the Kalman filter provides analytical expressions for posterior filtering. However, for non-linear models and non-Gaussian noise, such closed form expressions are almost impossible to obtain, and sequential monte carlo method provides its approximation. The basic idea of this method is to produce particles from the posterior densities, and these weighted samples provide approximations to the densities.In this dissertation, the applications of particle filter method in the civil engineering are mainly investigated. Firstly, its basic idea, method and improved method are introduced, then we exploit the specific steps and algorithm. Several kinds of common structural models are studied by numerical simulations in order to discuss its performance in structural identification of civil engineering models.On one hand, we apply the particle filter into the identification of several linear structural systems, and show that the particle filter method is effective. Furthermore, the influence of different noise level and the quantity of particles are experimented and some primary conclusions were obtained.On another hand, particle filter was introduced to the identification of Bouc-Wen Model which used to predict the non-linear behavior of many structural systems and the results was compared with EKF method. The particle filter simulation results show that in Gaussian environment both extended Kalman filter(EKF) and particle filter have almost the same tracking accuracy, and that in non-Gaussian noise environment particle filter has also good accuracy, while the...