Discrete-time high order neural control. Trained with Kalman filtering (Q932084)

The authors propose a new approach to nonlinear discrete-time tracking systems design based on control algorithms which use high order neural networks in order to ensure robustness to uncertainties and disturbances. Four different design techniques are proposed: {\parindent=4mm \begin{itemize}\item[{\(\bullet\)}] In the first method a high order neural network trained by an extended Kalman filter is used to solve the tracking problem for a class of MIMO discrete-time nonlinear systems with measurable state variables. \item[{\(\bullet\)}] In the second method a similar network is used to identify a model for such systems which is, in turn applied in design procedure for the block control approach and sliding mode technique. \item[{\(\bullet\)}] In the third method a recurrent high order neural network trained by an extended Kalman filter is used to design a Luenberger-type observer for a similar class of systems with measurable outputs and a controller is designed by means of the backstepping technique. \item[{\(\bullet\)}] In the fourth method the block control and the sliding mode technique are used while the state of the system is reconstructed using the same observer as in the third method. \end{itemize}} Stability analysis is based on the Lyapunov approach nevertheless its correctness is highly dependent on the separation principle which generally is not true for nonlinear systems. To illustrate efficiency of the proposed design methodologies the authors present results of their experiments with a benchmark which includes a PC for supervising, a PWM unit for the power stage, a dSPACE, DS 104 board for data acquisition and control of the system, and three phase induction motor as the plant to be controlled. All four design schemes are tested and the robustness of the control systems with respect to motor parameter variations and the load torque is demonstrated. The main idea of the book is to use the high order neural networks trained by the extended Kalman filter for neural identification, state estimation, backstepping control and block control with sliding mode and to combine these techniques in robust control design procedure for nonlinear discrete-time systems. The authors do not discuss problems related to discretization/sampling and control/estimation separation properties.