Adaptive fuzzy CMAC-based nonlinear control with dynamic memory architecture

An adaptive fuzzy cerebellar model articulation controller-based (CMAC) nonlinear control with the advantage of architecture learning is proposed. To cope with the tradeoff between the complexity of CMAC architecture and the quality of system convergence, a dynamic architecture learning scheme is introduced, where the associative memory reinforcement and the associative memory reorganization are involved. In the memory reinforcement process, new associative memories will be generated when the memory cells in the current architecture are found insufficient. On the other hand, the inefficient memories will be detected and reorganized in the memory reorganization process. With the proposed approach, the task of fuzzy CMAC architecture determination by preliminary knowledge or trials can be freed when a well-organized and well-parameterized CMAC is represented to achieve desired approximation performance. Thus, with the proposed CMAC, a dynamic control approach is presented. In this paper, according to the adaptive control theory, the fuzzy CMAC (FCMAC) is utilized as the main controller to mimic the ideal computation controller and a supervisory controller is designed to compensate the approximation error. In the FCMAC, all the controller parameters are online tuned based on the Lyapunov stability theorem such that the stability of closed-loop system can be guaranteed. Simulation results and comparisons are presented for verification.