Numerical Investigation of Gaussian Filters with a Combined Type Bayesian Filter for Nonlinear State Estimation

TY - GEN

T1 - Numerical Investigation of Gaussian Filters with a Combined Type Bayesian Filter for Nonlinear State Estimation

AU - Mehndiratta, Mohit

AU - Kayacan, Erdal

PY - 2016/8

Y1 - 2016/8

N2 - This study presents a numerical comparison of three filtering techniques for a nonlinear state estimation problem. We consider an Extended Kalman Filter (EKF), an Unscented Kalman Filter (UKF) and a combined type of Particle Filter, so-called Extended Particle Filter (EPF), for the state estimation for a re-entry vehicle system. The challenge in state estimation for this system is presence of significant nonlinearities in the process and measurement models. The performance aspects for the comparison include computation time, simulation time step, and effect of the choice of the initial conditions for the state estimate and covariance. Also, an investigation of the effect of the number of particles for EPF is performed. Simulation results illustrate that although EPF is computationally more expensive than EKF and UKF, it is less affected by the choice of initial conditions and simulation time step size.

AB - This study presents a numerical comparison of three filtering techniques for a nonlinear state estimation problem. We consider an Extended Kalman Filter (EKF), an Unscented Kalman Filter (UKF) and a combined type of Particle Filter, so-called Extended Particle Filter (EPF), for the state estimation for a re-entry vehicle system. The challenge in state estimation for this system is presence of significant nonlinearities in the process and measurement models. The performance aspects for the comparison include computation time, simulation time step, and effect of the choice of the initial conditions for the state estimate and covariance. Also, an investigation of the effect of the number of particles for EPF is performed. Simulation results illustrate that although EPF is computationally more expensive than EKF and UKF, it is less affected by the choice of initial conditions and simulation time step size.

KW - Bayesian filter

KW - Gaussian filter

KW - Kalman filter

KW - Nonlinear estimation

KW - particle filter

KW - unscented transformation

UR - http://www.scopus.com/inward/record.url?scp=85009216357&partnerID=8YFLogxK

U2 - 10.1016/j.ifacol.2016.10.206

DO - 10.1016/j.ifacol.2016.10.206

M3 - Article in proceedings

VL - 49

SP - 446

EP - 453

BT - IFAC-PapersOnLine

PB - Elsevier

T2 - 10th IFAC Symposium on Nonlinear Control Systems

Y2 - 23 August 2016 through 25 August 2016

ER -