An optimal tracking controller for nonlinear dynamic systems using multilayer neural networks

Feedforward neural networks are used to solve an optimal tracking control problem for discrete-time nonlinear dynamic systems with quadratic cost function. Control input to the plant is separated into two parts. One, called feedforward input, corresponds to the steady-state output of the plant. The other, called feedback input, corresponds to the transient-state output of the plant. Two multilayer neural networks are constructed as the feedforward and the feedback controllers. The feedback controller is trained by the backpropagation through time (BTT) algorithm to minimize a general quadratic cost function. The proposed methodology is useful as an off-line control method where the plant is first identified and then a controller is designed for it.