Machine learning-based routing in UAV-assisted telecommunication systems

DOI: 10.31673/2412-9070.2025.049695

Abstract

This article explores the prospects for implementing machine learning methods in the routing process in telecommunication systems with UAV integration. The existing routing protocols and their applicability to FANET systems are analyzed, in particular, taking into account the parameters of UAV node mobility, unstable communication and highly dynamic network topology. The main factors influencing the UAV-BS routing process and the prospects for integrating supervised learning, unsupervised learning and reinforcement learning are considered. The greatest obstacle to the deployment of machine learning-based routing strategies in FANET is the dynamic and unpredictable nature of the environment associated with fast movement and high mobility of air nodes. Machine learning algorithms should ensure high stability and scalability of the system in various operating conditions, including unstable connection quality and make a decision on routing with low latency. Stable and efficient operation of the routing protocol will be complicated primarily by frequent changes in the network topology, so the algorithm must be able to constantly adapt to changes and learn from huge amounts of temporary data when implementing routing decisions. The article considers routing protocols for telecommunication systems with UAV integration and explores the prospects for integrating machine learning methods into the routing process. It is determined that the only applicable machine learning method for integration into UAV systems is Supervised Learning, since this is the only type of machine learning capable of building a model without a specified data set. The article presents a method for integrating reinforcement learning into the routing process, taking into account four main characteristics when determining rewards for the learning agent.

Keywords: UAV; machine learning; artificial intelligence; routing; method; topology; reinforce ment learning; parameter; scalability; network.