Neural network approach to direct parameter adaptation of longitudinal autopilots

Abstract

An improvement of longitudinal autopilots consisting of the digital PI and P controllers is addressed in this paper. In order to achieve a good performance of these autopilots a direct adaptation of their three parameters is proposed. To this end, the two-circuit feedback is added by the feedforward circuit containing a neural network which needs to be trained offline. The input signals of this neural network correspond to the airspeed and the altitude of an aircraft whereas its output signals are the three controller parameters to be adjusted if flight regime changes. The behavior of a new longitudinal autopilot is studied by simulation experiments.