5G networks present new possibilities in communication technology, but they also create challenges in network management due to the incorporation of new complex concepts such as network slicing and virtual network functions (VNF). These challenges make it difficult for network operators to manually ensure that all quality of service (QoS) requirements are met across all network slices while also monitoring resource and energy consumption. To address this, automated network assurance solutions are required. Although machine learning (ML) and deep learning (DL) techniques have shown potential in this field, they come with difficulties such as acquiring real-world labeled training datasets and guaranteeing the quality of ML/DL pipelines during production.
The thesis proposes a time-driven closed-loop algorithm with proactive components that are differentiated by slice type, key performance indicator (KPI) type, and current resource consumption levels to maintain QoS for 5G end-to-end (E2E) network slices. Through simulations, we show that the proposed closed-loop algorithm is effective in resolving KPI violations and reducing network and compute resource usage, as well as allowing for flexible tradeoffs between QoS guarantees and resource consumption for each slice type via slice-specific parameter adjustments. Our solution not only enables network providers to better negotiate with customers, but also has the potential to generate training data for future ML/DL approaches.