Title: Extended Kalman filter steady gain scheduling using k-means clustering
Authors: Jun Chen
Addresses: Department of Electrical and Computer Engineering, Oakland University, Rochester, MI 48309, USA
Abstract: This paper studies the gain scheduling problem for extended Kalman filter (EKF). To save throughput, the steady gain is usually used for Kalman Filter. In the context of EKF, there is no universal steady gain. In this paper, we propose a methodology to schedule the steady gain for EKF. The idea is to offline linearise the nonlinear model at various operating points, and for each of the linearised systems, to compute the steady gain corresponding to conventional Kalman filter by solving the corresponding algebraic Riccati equation. The operating space is then divided into multiple zones, through k-means clustering algorithm, so that within zone, the steady state gains are close to each other. For real time filtering, the centroid of each zone is used as gain for correction step, instead of computing the time-varying gain online, hence saving throughput. We demonstrate the proposed methodology in a two-state nonlinear system.
Keywords: state estimation; EKF; extended Kalman filter; steady gain; k-means; gain scheduling; throughput; embedded computation.
DOI: 10.1504/IJMIC.2020.110356
International Journal of Modelling, Identification and Control, 2020 Vol.34 No.2, pp.158 - 162
Received: 27 Sep 2019
Accepted: 09 Feb 2020
Published online: 15 Oct 2020 *