Adaptive control of nonlinear dynamic systems using \(\theta\)-adaptive neural networks (Q1129706)

For dynamic systems with nonlinear parameterization, the adaptive control problem is addressed by designing a neural-network-based controller with a particular architecture. The structure of the controller is derived from the available physical model, whereas the parameters are estimated by a neural network. It is shown that the proposed adaptive controller ensures the closed-loop stability of the overall system. The neural network training is approached as an optimization problem with constraints representing sufficient conditions for the closed-loop stability. This constrained problem is eventually reduced to an unconstrained one, which can be solved by various optimization methods. Simulation experiments allow a comparison of the results obtained by the procedure proposed in this paper with other methods.