Analysis of System Identification Using the Neural Extended Kalman Filter

Abstract

The neural extended Kalman filter (NEKF) is an on-line state estimation technique that can be used to identify differences between the a priori mathematical model and the actual system dynamics. Several applications of this algorithm have been applied to control problems. Two of the approaches seem most promising. The first is to place the NEKF in the feedback loop as an adaptive state estimator. The second is to use the NEKF as a system identification scheme outside the control loop and periodically replacing the model used by the state estimator with that of the NEKF. These two approaches are compared to determine the benefits of each by applying them to an example control system where the system model used in the control law was a mismodeled version of the actual system.