Programmable linear-quadratic regulator

A feedforward neural network is used to adjust LQR gains in the case of non-stationary state matrix.

This model demonstrates that the LQR design approach can be effectively used also for plants characterized by a non-stationary state matrix. A set of controllers is designed for different working points and an FFNN is employed to store this knowledge. The gain matrix K is then adjusted on the fly if the state matrix A changes. In the case of induction motor some entries of the state matrix A are functions of e.g. the stator flux space vector angular velocity and the slip angular velocity of the rotor. More details can be found in Neural-Network-based Programmable State Feedback Controller for Induction Motor Drive (http://dx.doi.org/10.1109/IJCNN.2006.246811). It is also advisable to get familiar with LQR basics prior to playing with this model. See e.g. http://www.mathworks.com/help/control/ref/lqr.html .