Online parameter and process covariance estimation using adaptive EKF and SRCuKF approaches

Two observers for joint parameter and state estimation are presented in this paper. The observers are based on the Extended Kalman Filter (EKF) or the Square Root Cubature Kalman Filter (SRCuKF) and a Recursive Predictive Error (RPE) method for state and parameter estimation, respectively. Sensitivity models are introduced to compute and minimize a cost functional and then recursively estimate parameter and process covariance values online. The algorithm performance is tested using simulation models of two test benches. Simulation results show that the novel method based on SRCuKF is more accurate than the adaptive EKF and gives improved results with stiff and highly nonlinear systems. A projection algorithm and an adaptive gain for the RPE are introduced to make the complete observer more stable.