Comparativo entre controlador PID e redes neurais artificiais no controle de atitude de um quadricóptero

Abstract

A quadcopter is a helicopter with four Motors, that make the vehicle more stable but more complex to control. A quadcopter has six degrees of freedom, three of them regarding the position: height, horizontal and vertical motions; and the other three are related to the orientation: pitch, roll and yaw. This work presents a study of using Artificial Neural Networks and PID control applied to the attitude control of a real quadcopter beyond conception of four Neural Networks using algorithms Backpropagation, Levenberg-Marquardt, Bayesian Regularization and Scaled Conjugate Gradient with supervisioned learning by PD control, making a control systems named here as PDNN, for this control systems was summed the Integral parcel of PID traditional control, resulting in four hybrid control systems called PDNN+I, of which for the case of this work, the Neural Network generated by the training with the algorithm Backpropagation, Levenberg-Marquardt and Bayesian Regularization obtained a better result, resembling to its teachers.