Point-To-Point Iterative Learning Control Based On Neural Networks Approximation

After iterative learning control was first proposed, it has attracted intensive research interests. In last decades, both theory and application of ILC has been intensively developed. ILC is basically a data-driven based learning algorithm which shows effectiveness in dealing with repetitive control tasks for batch process. The objective of conventional ILCs is to track a desired output trajectory in a given time interval. However, maybe only some critical points of output need to be considered accurately. In some special scenarios, only the terminal output is needed to be controlled:(1) the only available measurement is the terminal state or terminal output, so the tracking error in entire operation interval is not available; (2) the ultimate control objective is also the terminal state or terminal output instead of the entire trajectory. Terminal iterative learning control (TILC) is introduced to deal with this unconventional control task. Considering more general case, point-to-point iterative learning control (P2PILC) is designed for tracking several critical points of output. Both terminal iterative learning control and point-to-point iterative learning control are considered in this thesis. In order to avoid the difficulty of finding solution for nonlinear systems, radial basis function (RBF) neural network is utilized to approximate the input of system. A dead-zone like auxiliary error function is introduced to compensate the approximation error of RBF neural network. Additionally, both fixed input case and time-varying input case are discussed to demonstrate the performance of algorithm. Then the technical convergence analysis is provided by Lyapunov like function. Finally, the proposed TILC is successfully applied to train station stop control and batch reactor. Furthermore, a numerical example is provided to show the effectiveness proposed P2PILC algorithm.