Robust control of a linear system with plant uncertainties using an artificial neural network

This dissertation investigates the design of a robust controller for a linear system based on an Artificial Neural Network (ANN) and compares that approach with conventional methods. LQR (Linear Quadratic Regulator), LQG (Linear Quadratic Gaussian), H{dollar}sbinfty{dollar}-constrained LQR control, and H{dollar}sbinfty{dollar}-constrained LQG control problems are considered. Each controller design technique is explained in terms of robustness in the frequency domain.; The dynamic system is assumed to be in one of a finite number of configurations, each of which contains a bounded plant uncertainty, corresponding to which exists a pre-designed stabilizing controller. Multi-layer neural networks are used for the indentification of a plant uncertainty level which allows us to choose the best controller among a finite number of given controllers. Simulation results are given in every chapter. The examples given in Chapters 7 and 8 especially reveal that an efficient robust controller design using ANN can be achieved.