Application of the free-model based neural networks in model reference adaptive inverse control

The free-model concept is implemented in the design of neuro-identifier and neuro-controller in the model reference adaptive inverse control. Through the off-line training of both neural networks, the proposed scheme obtains satisfactory performance on the tracking, pole positioning, and disturbance rejection in the inverted pendulum problem. Besides, it is shown that the control scheme implemented by the free-model based neural network can obtain more satisfactory performance than the one implemented by the conventional neural network. The conditions of satisfactory free-model approximation errors are derived for different cases. The implementation of the backward difference operator and inputs for the free-model based neural network are discussed.