A new approach for adaptive control of a nonlinear system using neural networks

An approach for controlling a nonlinear system using an adaptive scheme implemented by neural networks is presented. Two neural networks are trained in three different stages. In the first stage, the nonlinear system is excited several times to teach the inverse dynamics of the system to a neural network. In the second stage, the system is again excited several times to train a second neural network with input signals that will control the nonlinear system in the desired way. After the first two stages of training, the system is operated with the second neural network as feedback, and its weights are adaptively adjusted to accommodate possible parameter variations in the nonlinear system. Results obtained with a simulation program developed to train the neural networks using the backpropagation algorithm and input-output-state data are presented.