The dissertation is devoted to the solution of the current scientific and technical task of developing a methodology for building and using machine learning (ML) models for mobile agents that act as unmanned aerial vehicles (UAVs), using an automated pipeline based on Kubeflow using the best practices of the ML concept and MLOps (Machine Learning and Operation) processes. A strategy for using these models for mobile agents using Mesh networks, which is part of the described pipeline, is developed. The development of modern information technologies as data science, data analytics and ML are becoming one of the main tools for solving complex applied problems in various spheres of activity. ML is one of the methods of Artificial Intelligence (AI), in particular, the practical implementation of its capabilities by creating algorithms for identifying patterns during the analysis of big data, and their further use for self-learning. This issue is especially relevant in the context of the study of UAV warfare with the help of AI as an option for giving the military a tactical advantage over the enemy and reducing the risk to human lives. Using ML with AI algorithms, UAVs can quickly identify threats, respond to them, conduct surveillance and reconnaissance. Considering the above, conducting research on the development of a new method with the use of integrated ML and AI in controlling the combat tactics of UAVs that is resistant to radio-electronic interference is currently an urgent task. To achieve the goal set in the work, it is necessary to solve the following separate research tasks: 1. to analyze the current state of development of ML and AI with the possibility of their application in UAVs. 2. investigate the possibilities of using modern MLOps solutions to improve the development processes of information systems of the ML in the context of Kubeflow tools. 3. develop a model for building and applying ML with the possibility of continuous training, which is achieved by a high level of integration and automation of the pipeline using Kubeflow components and the Kubernetes platform. 4. Experimentally check the results of operating the model in the Kubeflow arsenal using such improvement factors as speed of development, implementation of changes, reduction of time to search for problems, recovery after global interruptions, reduction of the number of errors in the model. 5. Develop an architectural concept of the system based on the Kubernetes k3s distribution using a machine learning model in the edge computing paradigm for mobile agents using mesh networks. 6. to evaluate the effectiveness of using a wireless Mesh network to increase the functional stability of the distributed information system of the of the UAV. Scientific novelty of the obtained results. In the process of theoretical and experimental research and modeling, the following new scientific results were obtained: 1. The concept of a full-fledged information solution based on a continuous integration pipeline with the possibility of continuous training, a high level of integration and automation using Kubeflow components and the Kubernetes platform was formed, which allows to improve the quantitative and qualitative components of experiments, reduce the time spent on its preparation, and minimize errors caused by the human factor. 2. The architectural concept of the system based on the Kubernetes k3s distribution was designed, which allows the effective use of ML models in the paradigm of edge computing (Edge Computing) with the use of Mesh-networks for mobile agents (UAVs), which form a functionally stable distributed information system. 3. The model for increasing the functional stability of the distributed information system of UAVs using a wireless Mesh-network of data transmission has been improved, which will allow to resist radio-electronic and other interferences and work in autonomous or semi-autonomous mode.
The practical significance of the obtained results in the field of development and creation of effective methods of building and using ML models for UAVs is as follows: 1. The proposed architectural solutions of information systems and the method can be used by research and development organizations and state structures of the Armed Forces of Ukraine to implement the concept “Analytical decision support system for UAVs”, which is shown through functional process models (AS, TO-BE and SHOULD- BE). 2. Modern MLOps solutions based on Kubeflow tools for the creation and use of educational models of ML are implemented in the discipline “Decision support systems”. A high level of automation and integration of components allows you to create and run a test model in a relatively short time, but without reducing the quality of its work and reliability.
Key words: distributed information systems, machine learning, artificial intelligence, mobile agents, Mesh networks, unmanned aerial vehicle, MLOps, Kuberentes, Kubeflow, Edge Computing
