CMAC neural networks for control of nonlinear dynamical systems: Structure, stability and passivity (Q1361317)

The cerebellar model articulation computer (CMAC) neural networks are used for controlling multi-input multi-output nonlinear systems in canonical form. It is shown that the CMAC neural network uniformly approximates all the nonlinearities in the system and, consequently, guarantees tracking performance. The weight-tuning algorithm relies on a modification of the classical procedures, which is able to overcome the requirement of the persistency-of-excitation condition. The robust stability of the closed-loop system is derived from the passivity properties. A simulation example illustrates the theoretical concepts developed in this paper, and allows the comparison with a standard adaptive controller.