HYBRID METHOD OF ADAPTIVE CONTROL OF VARIABLE MODE OF UNMANNED AERIAL VEHICLES WITH INTELLIGENT ONLINE COMPENSATION OF DISTURBANCES

Abstract

The article resolves the current scientific and technical contradiction between the need to increase the accuracy of navigation control of autonomous unmanned aerial vehicles (UAVs) and the strict resource constraints of on-board computing systems. An intelligent-robust control architecture is proposed, based on the synthesis of the adaptive alternating mode method (ASMC) and the recurrent neuro-fuzzy network RSEFNN. The scientific novelty of the work lies in the improvement of the hybrid approach, which, unlike classical robust methods, uses an intelligent observer for online identification and compensation of nonlinear components of dynamics and external disturbances. This made it possible to significantly reduce the gain coefficients of the discontinuous part of the controller, minimize the "rattling" effect, and increase the energy efficiency of actuators. Mathematical proof of the stability of the closed-loop system using the direct Lyapunov method confirmed the asymptotic convergence of trajectory tracking errors to zero and guaranteed the numerical stability of the neural network training processes. An important practical contribution is the implementation of methods for suppressing high-frequency oscillations by replacing the discontinuous control function with its smooth approximation based on the adaptive boundary layer and hyperbolic tangent. To ensure the determinism of the computational cycle in real time, optimization using the Padé method was applied, which allowed minimizing algorithmic latency and achieving a control frequency of up to 1000 Hz on embedded CPUs without specialized accelerators. The results of the comparative analysis confirmed the high robustness of the developed method under conditions of intense wind loads. In particular, the use of the ASMC+RSEFNN controller allowed to increase the positioning accuracy in steady state by 10.2–12.6 times compared to classical PID controllers. The integrated neuro-fuzzy identifier provided effective compensation for systematic wind shear, which is a critical factor for performing UAV precision guidance tasks in difficult meteorological conditions.