(A) study on the design of robust learning controller based on a new learning paradigm

Abstract

Perturbations such as state disturbances, measurement noise, initialization error and set-point changes are inevitable in learning control systems and tend to gradually deteriorate the performance of the learning control systems. To resolve this problem, when the perturbations are introduced in the learning control systems, the behavior of the learning law is investigated in the error state space and a novel learning paradigm that can efficiently eliminate the effect of perturbations on the overall system is proposed. The proposed paradigm of learning says that the learning or rule modification in the controller should occur in consideration of both the system``s current performance and the system``s behavioral tendency as well. At first, it is considered to design robust direct and indirect fuzzy logic-based learning controllers that can be applied for tracking control of a class of uncertain nonlinear SISO systems. It is shown that, in the presence of the perturbations such as approximation error of fuzzy system and external signals, boundedness of all the signals in the system is ensured and, while under the assumption of no perturbations, the stability of overall system is guaranteed. The concept of persistent excitation in the fuzzy logic-based learning control systems is first utilized to guarantee the convergence and boundedness of adaptation parameter in the proposed learning controllers. Robustness and stability of the proposed control system are proved. Also, it is observed that parameter and tracking error convergence in the proposed method can be smoother and faster than the previous fuzzy learning controllers. Two simulations are conducted to show the robustness property of the proposed controllers. Then, a robust self-learning fuzzy controller for a class of nonlinear MIMO systems is proposed based on the proposed learning paradigm. The well-known techniques of sliding mode control and the fuzzy decision making method are utilized to implement t...