Neural network based robust nonlinear control

Online trained neural networks have become popular in recent years in the design of robust and adaptive controllers for dynamic systems with uncertainties due to their universal function approximation capabilities. This research explores the application of online neural networks for the design of model following controllers and for dynamic reoptimization of a Single Network Adaptive Critic (SNAC) optimal controller. Model following controllers for a general class of nonlinear systems with unknown uncertainties in their modeling equations have been developed in this research. A desirable characteristic of the model following controller scheme elaborated in this work is that it can be used in conjunction with any known control design technique. This research also discusses a technique that dynamically re-optimizes a Single Network Adaptive Critic controller. The SNAC based optimal controller designed for the nominal plant model no more retains optimality in the presence of uncertainties/unmodeled dynamics that may creep up in the system equations during operation. This necessitates the application of online function approximating neural networks that can help in SNAC reoptimization. Neural network weight update rules for continuous and discrete time systems have been derived using Lyapunov theory that guarantees both the stability of error dynamics and boundedness of the neural network weights. Detailed proofs and numerical simulations of the online weight update rules on various engineering problems have been provided in this document.