Towards Autonomous Networks: Creating and Orchestrating Intelligence for Next-Generation Network Management

Softwarization and Virtualization have shaped modern 5G networks, transforming rigid network architectures into flexible, service-based infrastructures, enabling dynamic reconfiguration, scalability, and automation. Despite the tangible benefits of reducing CAPEX and OPEX and increasing resource efficiency, these two trends exacerbate the complexity of network management, requiring sophisticated orchestration solutions that minimize human intervention. While traditional approaches for network management enable automation, AI and ML have emerged as powerful tools for network operations by leveraging vast amounts of data to enable higher levels of autonomy in network management. The application of AI and ML in networks creates what we call Network Intelligence (NI), allowing network functions to adapt dynamically, predict network changes, and automate decision-making through supervised learning, reinforcement learning, and unsupervised learning techniques. The shift towards next-generation networks calls for a deeper fusion of network programmability with AI, paving the way for fully autonomous management. While standardization bodies have begun integrating NI into network architectures, current solutions remain limited. Existing network designs do not fully accommodate NI, and research has primarily focused on isolated NI models rather than achieving end-to-end integration. Effective coordination among multiple NI functions is crucial for global optimization and network stability. This dissertation addresses these challenges by exploring the design and orchestration of AI-driven NI for next-generation networks. A novel NI Stratum (NIStratum) is proposed, supported by an end-to-end NI Orchestrator (NIO) for managing NI lifecycles across different network domains. The architecture ensures scalable deployment, conflict resolution, and efficient data management, reusing elements from known network management frameworks and guaranteeing its integration with current standards. Two NI solutions are proposed: (i) an NI model for Wi-Fi performance prediction based on Graph Neural Networks (GNNs) and a coexistence management solution that leverages a technology-agnostic spectrum recognition method that uses Convolutional Neural Networks (CNNs), and (ii) an NI-based VNF autoscaler using Deep Reinforcement Learning (DRL), that autonomously learns the operation thresholds to keep two contrasting objectives under control. The dissertation also introduces a methodology to evaluate DRL performance under different conditions, highlighting the challenges of real-world AI integration in network management. This doctoral thesis advances autonomous network management by integrating AI/ML with next-generation architectures, proposing intelligent frameworks for continuous adaptation, enhancing efficiency and scalability, and driving innovation toward fully autonomous networks.

• Autores: Soto Arenas, Paola Andrea
• Tipo de recurso: Doctoral thesis
• Fecha de publicación: 2025
• Institución: Universidad de Antioquia
• Repositorio: Repositorio UdeA
• Idioma: eng
• OAI Identifier: oai:bibliotecadigital.udea.edu.co:10495/47606
• Acceso en línea: https://hdl.handle.net/10495/47606
• Palabra clave: Software-defined networking (Computer network technology); Aprendizaje automático (inteligencia artificial); Machine learning; Administración de redes de computadores; Network management; Redes de computadores; Computer networks; Autonomous Networks; http://id.loc.gov/authorities/subjects/sh2014000092; ODS 9: Industria, innovación e infraestructura. Construir infraestructuras resilientes, promover la industrialización inclusiva y sostenible y fomentar la innovación
• Rights: openAccess
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/