Ідентифікація математичної моделі лабораторної установки ланки квадрокоптера

Abstract

The paper describes a laboratory setup of a quadrocopter link, which is a nonlinear plant. Experimental studies on the control of the movement of the installation were carried out and an array of experimental data was obtained. In order to perform the identification of the mathematical model of the installation, the array was processed. The procedure is consisted of separate stages. At the beginning of the calculations, measurement errors were eliminated. Subsequently, an array of discrete values of the angular velocity of the rod (quadrocopter link) movement is determined. After that, the resulting array was filtered and data was generated in a format suitable for training an artificial neural network. Such data included the pairs: „current value of the voltage drive, current value of the angle, current value of the angular velocity” - „subsequent value of the angle, subsequent value of the angular velocity”. The neural network (predictor) was a single-layer feedforward network with three inputs and two outputs. Artificial neural network has been trained according to the paradigm of supervised training. As a result, a predictor has been obtained that allows predicting the behavior of the plant under a certain control (drive supply voltage). The quality of the predictor's work was estimated based on the analysis of graphical dependencies and in terms of the standard deviations of the experimental (in the case of angular velocity – calculated) and predicted values. It gives grounds to state that the obtained predictor (mathematical model of the plant) may be used in order to synthesize control systems.