System Identification using the Neural-Extended Kalman Filter for Control Modification

Abstract

The neural extended Kalman filter has been shown to be able to work and train on-line in a control loop and as a state estimator for maneuver target tracking. Often, however, the design of a control system does not have a state estimator in the feedback loop. The ability of the NEKF to learn dynamics in an open-loop implementation, such as with target tracking and intercept prediction, can be used to identify mis-modeled dynamics. The improved system model can then be used to adapt the control law to provide better performance based on the actual system dynamics. This new approach to neural extended Kalman filter control operations is introduced in this work using applications to the nonlinear version of the standard cart-pendulum system.

Robust GMDH-type neural network with unscented Kalman filter for non-linear systems

“…The BP algorithm has some drawbacks, e.g. depending on how the primary weight and the number of hidden neurons are chosen, as well as its low convergence speed, it is very sensitive to the noises present in the data sets used to train, and, therefore, it has poor generalization in the modelling of complex non-linear functions (Nariman-zadeh et al, 2003;Stubberud, 2006;Yang et al, 2007).…”