Artificial neural network feedback loop with on-line training

In this paper, we discuss the implementation of a neural network feedback loop, a state estimator and a state feedback controller, that trains online to overcome variations between the a priori model dynamics and the true system dynamics. The technique uses the previously developed neuro-observer, an extended Kalman filter augmented with a neural network, and a model reference adaptive control law. Both neural networks, the state estimator's and the controller's, are trained using extended Kalman filter training paradigms. The paper also includes a discussion of parameter selections for the neural networks and their training paradigms.