Neural control of nonlinear systems with composite adaptation for improved convergence of Gaussian networks

The use of composite adaptive laws for control of the affine class of nonlinear systems having unknown dynamics is proposed. These dynamics are approximated by Gaussian radial basis function neural networks whose parameters are updated by a composite law that is driven by both tracking and estimation errors. This is motivated by the need to improve the speed of convergence of the unknown parameters, hence resulting in better system performance. To ensure global stability despite the inevitable network approximation errors, the control law is augmented with a low gain sliding mode component and deadzone adaptation is used for the indirect part of the composite law. The stability of the system is analyzed and the effectiveness of the method is demonstrated by simulation.