Improving the dynamic response of neural network controllers using velocity reference feedback.

A neural network controller designed solely based on the desired response of the dynamics to realize a maximum acceptable steady-state error may not be the best one in terms of the transient responses. It may have an excessive overshoot and larger risetime. Similarly, a controller having a very good transient response may not have the desired steady-state characteristics. The effectiveness of using a velocity reference feedback for improving the dynamic response of the neural network controllers is investigated. Using the direct control scheme, on an underwater vehicle with nonlinear dynamics, it is shown that specific improvements in overshoot, risetime, and settling time can be achieved using this method.