Abstract

The exponential growth in mobile data traffic and connected devices necessitates intelligent management solutions for 5G and B5G networks. Modern telecommunications networks must simultaneously support diverse services ranging from video streaming to autonomous vehicles, each requiring different performance guarantees. This thesis presents a comprehensive framework that uses artificial intelligence to automatically manage network resources and maintain service quality. We developed a closed-loop control system where reinforcement learning agents continuously monitor network performance, make resource allocation decisions, and learn from the consequences of those decisions as they ripple through the network. Each action taken by the system influences future network conditions, creating a complex feedback loop that traditional management approaches struggle to handle. Our framework learns to dynamically adjust computational resources (CPU, memory, storage) and network bandwidth in response to changing traffic patterns, preventing service degradation before it occurs. The system eliminates the need for manual tuning by automatically learning the delicate balance between maintaining service quality and minimizing resource consumption. We implement both single-agent and multi-agent architectures, where multiple independent agents can manage different network segments simultaneously. Validation using packet-level network simulator with traffic patterns derived from urban mobility data demonstrates that our approach successfully maintains performance metrics such as delay, packet-loss, and jitter within acceptable bounds while adapting to dynamic conditions. The multi-agent configuration achieves efficient resource management with reduced computational overhead. This work establishes foundational principles for autonomous network management, providing practical frameworks for deploying intelligent, adaptive, and energy-efficient control mechanisms in next-generation telecommunications infrastructure.