Stochastic system identification via filtering techniques

simultaneous tracking of the state and identification of hidden parameters for structural systems are usually pursued by recursive Bayesian inference schemes. Provided that the state space equations are linear and the distribution of uncertainties is Gaussian, the Kalman filter furnishes the optimal solution to the recursive Bayesian estimation problem. In practice, only few real problems can be assumed to evolve linearly; moreover, the probability density functions are seldom Gaussian. In this work, we compare the performances of sigma-point Kalman filtering and particle filtering for linear and highly nonlinear structural systems, so as to assess their stability and robustness in stochastic system identification.