ABSTRACT
It has been demonstrated by many researchers that an unknown dynamic plant can be made to track an input command signal if the plant is preceded by a controller which approximates the inverse of the plant’s transfer function. Precascading a plant with its inverse model provides an unity mapping between the input and output signal space. This concept of inverse modeling has been referred to as adaptive inverse control. However, the concept of transfer function is limited to linear systems, and the control algorithms developed under this framework can not be extended to nonlinear systems. Due to the functional approximation and learning capabilities, the artificial neural networks can be employed to extend the concept of adaptive inverse control to nonlinear systems. In this paper, two dynamic neural structures, called recurrent neural network and dynamic neural processor, are used to coerce the nonlinear systems to follow the desired trajectories based on the principle of adaptive inverse control. In the process, we compare the performance of these two dynamic neural structures as applied to the inverse control problems.
