Switched learning adaptive neuro-control strategy

Abstract

The generalized learning algorithm can be efficiently used as control strategy, but it has some drawbacks such as: sensitivity to the training dataset, poor robustness against changes in the system, difficulty to generate the control signals without destabilising the plant, tuning of the controller, etc. To overcome some of these issues, in this work a new switched neural adaptive control strategy is proposed. It is based on the combination of an adaptive artificial neural network, a PID regulator, an estimated inverse model of the plant and two switches to route the signals properly in the control scheme. The technique is described using the hybrid automata formalism. In order to test the validity of this proposal, it is applied to the control of a quadrotor unmanned aerial vehicle (UAV), subjected to changes in its mass and wind disturbances. Simulation results show how the on-line learning increases the robustness of the controller, reducing the effects of the mass change and of the wind on the UAV stabilization, thus improving the UAV trajectory tracking.