| The aim of this thesis is to show that in decentralized control configuration, learning control idea can be applied to achieve good trajectory tracking performance. In the first part of the present work, we provide some sufficient conditions for the convergence of decentralized learning control. We also investigate the problem of learning in noisy environment. We show that decentralized learning control can effectively reject noise signals to the extent that they are periodic. However, if the noise signals are non periodic but bounded, the tracking errors are shown to be confined within a bound that is inversely proportional to the rate of convergence of the learning algorithm. In the second part, we study the problem of robustness of the learning control to the interconnection perturbations (connective convergence). The main results are some sufficient conditions for the connective convergence of the decentralized learning control. In the last part of this work, we introduce problem of coordination among subsystems using the idea of task decomposition. By using least-square techniques, we show how a trajectory tracking problem can be decomposed into a set of subproblems prior to the application of decentralized learning control. |
