Increasingly, for many application areas, it is becoming important to include elements of nonlinearity and non-Gaussianity in order to model accurately the underlying dynamics of a physical system. The problem of identifying nonlinear system models arise in various applications in control and signal processing. In this context, one of the most successful and popular stastical identification approaches is Particle Filtering, otherwise known as Sequential Monte Carlo (SMC) methods. As compared to Extended Kalman Filter and Gaussian Sum Filter, this approach is computationally reliable for identification of highly nonlinear systems in terms of accuracy, and, at the same time chance of failure in difficult circumstances decreases. The numerical integration techniques, on the other hand, are only feasible in low-dimensional state-spaces. In this paper the particle filtering approach has been attempted for non-linear system identification. The particles and their associated importance weights in particle filtering approach evolve randomly in time according to a simulation-based rule. This is equivalent to a dynamic grid approximation of the target distributions, where the regions of higher probability are allocated proportionally more grid positions. Using these particles Monte Carlo estimates of the quantities of interest may be obtained, with the accuracy of these estimates being independent of the dimension of the state space. The envisioned method is easier to implement than classical numerical methods and allows complex nonlinear and non-Gaussian estimation problems to be solved efficiently in an online manner. The experimental results on comparison with Kalman filtering show the efficacy of the proposed method through illustrative examples.

Non-linear System Kalman Filter Bayesian Filter Sequential Estimation Particle Filter