METHOD OF SOFTWARE IMPLEMENTATION OF INTELLIGENT ALGORITHMS FOR CONTROL OF UNMANNED AERIAL VEHICLES IN HARD REAL-TIME SYSTEMS

Abstract

The article resolves the current scientific and technical contradiction between the need to increase the accuracy of precision guidance of unmanned aerial vehicles (UAVs) in conditions of intense interference and severe limitations of computing resources of onboard systems. The presented work focuses on the creation of an intellectually robust control architecture capable of ensuring stable functioning in conditions of time latency, noisy navigation data and dynamic uncertainty of the object. The scientific novelty of the research lies in the development and implementation of a recurrent self-evolving neuro-fuzzy network (RSEFNN), integrated into the adaptive switching mode controller (ASMC) circuit for online identification and compensation of non-stationary external disturbances. The key feature of the developed architecture is the combination of high robustness with computational efficiency, which is achieved through mathematical optimization of transcendental functions using the Padé method, which allowed to reduce resource consumption by six times compared to standard implementations. The use of self-evolving structures with a strict restriction on the number of rules guarantees determinism of execution time (WCET), which is critically important for aviation certification of on-board software. An important practical result was the creation of a software simulation bench in the MATLAB environment based on an object-oriented approach and a fixed integration step of the 4th-order Runge-Kutt method, which ensures full reproducibility of numerical experiments. The developed algorithm for deterministic actuation distribution allows to effectively control a UAV with an excess number of engines without using iterative procedures, guaranteeing a constant execution time of operations regardless of the input signals. Experimental validation using the Dryden spectral turbulence model confirmed the high robustness of the hybrid system under conditions of intense stochastic disturbances typical of low altitudes. Statistical profiling showed that the ASMC+RSEFNN method consumes less than 2 KB of RAM and has a runtime margin of more than 33% relative to the critical limit of 2 ms. Compared to deep learning and nonlinear predictive control methods, this approach demonstrates significantly higher computational efficiency, allowing to combine intelligent noise compensation with strict real-time requirements on ARM Cortex-M7 microcontrollers.