Trajectory Control Method Of UAV In Autonomous Flight Mode Using Neural Network MELM Algorithm

• Published in: 2020 IEEE 2nd International Conference on Advanced Trends in Information Theory (ATIT) pp. 114 - 118
• Main Authors: Fesenko, Oleksii, Bieliakov, Robert, Radzivilov, Hrygorii, Hulii, Volodymyr, Kovalchuk, Oleh, Korotchenko, Liudmyla
• Format: Conference Proceeding
• Language: English
• Published: IEEE 25.11.2020
• Subjects: Biological neural networks; Global Positioning System; Heuristic algorithms; Neurons; Prediction algorithms; Quaternions; Recurrent neural networks; Training; Trajectory; Unmanned aerial vehicles
• DOI: 10.1109/ATIT50783.2020.9349317

in our time, the use of unmanned aerial vehicles (UAVs) for communication purposes, for example in Flying Ad-Hoc Networks, often correlates with the emergence of a number of scientific and technical tasks for designing flight control systems of UAVs in the absence of GPS signals: Firstly, in connection with the requirement for the number of errors of setting the flight path. Secondly,compliance with computational accuracy requirements with power limitations of the microcomputer base. In order to solve these problems, it is proposed to use the Method of trajectory control UAV in autonomous flight mode using neural network. Multiple Hidden Layers Extreme Learning Machine (MELM) algorithm, the result of which is tested using experimental comparison with existing Extreme learning machine (ELM) algorithms - Kalman and Weight Agnostic Neural Networks (WANN) - Recurrent Neural Network (RNN) Madgwick. To solve the scientific and technical problem, for the first time it was proposed to use a unit for converting navigation data into a quaternion form to reduce the dimension of the input data, which in turn made it possible to increase the speed and accuracy of training the neural network without using the quantization process. For the first time, the MELM neural network algorithm was applied to solve autonomous navigation problems to reduce the deviation of UAVs from the target trajectory during the disappearance of GPS signals. It has been experimentally proved that the application of the developed technique has ensured a relatively slightest error in predicting the navigation parameters of the trajectory, which occurs in the event of the disappearance of signals of global satellite positioning systems.