A Feedback Scheme for Improving Dynamic Response of Neuro-Controllers

A neural network controller designed solely to realize maximum acceptable steady state error may have unsatisfactory transient response such as excessive overshoot and large rise time. Conversely, a controller designed to realize very good transient response may have unacceptable steady state characteristic. In this paper we investigate the effectiveness of adding velocity reference feedback for simultaneously improving the transient and steady state response of a neural network controller. Using the direct control scheme on the nonlinear dynamics of an underwater vehicle, we show that specific improvements in overshoot, rise time and settling time could be achieved with this method. The control scheme and results are discussed in detail.