A methodology to reduce large overshoot in a direct neural control system

Direct neural control presented a good relearning performance in tracking a desired trajectory. However, random distribution of the initial weights of the neural network controller results in large initial overshoots. In this paper? a closed-loop training method for a direct neural controller is pro posed, aiming to generate 'good' initial weights for direct control. The tr aining data are generated on the on-line trajectory tracking using a conven tional control guide. Pre-training of the neural network concentrates on a subset that system states mainly fall in. The simulation studies for a sing le-link manipulator have verified that the trained direct neural control sy stem exhibits a better system response than an untrained neural control sys tem does in trajectory tracking and set-point regulation with significantly reduced initial overshoots.