Sustainable AI for Zero Touch Network & Service Management with Graph Neural Networks

In the pursuit of sustainable and efficient 6G network management, leveraging advanced AI methodologies becomes imperative. Our work introduces a comprehensive framework for resource usage prediction, focusing on data-compute 6G networks. This novel approach relies on state-of-the-art AI/ML techniques to achieve accurate resource usage and energy consumption predictions in 6G networks. By predicting various metrics obtained from Prometheus and Kepler, such as memory usage and energy consumption, we dynamically scale services across nodes using the Edge Platform of Intracom Telecom, aiming to reduce the energy footprint while preventing performance degradation and maintaining high Quality of Service (QoS). A core component of our approach are the Graph Neural Network (GNN) models, which are based on a Spatio-Temporal Graph Neural Network (STGNN), a very capable deep learning technique in predicting future values by modeling both spatial and temporal dependencies among network nodes. This enables a proactive and accurate forecasting mechanism that outperforms traditional Attention Temporal Graph Convolutional Network (A3T-GCN), but also cutting-edge Large Language Models (LLMs) in time series forecasting tasks. Our framework additionally incorporates automated AI models Life Cycle Management (LCM) using zero-touch MLOps techniques through tools like Kubeflow and Katib, ensuring efficient deployment, monitoring, and maintenance of AI models. Our extensive evaluation demonstrates the superiority of our approach in terms of prediction accuracy and operational efficiency compared to A3T-GCN and LLM based methods. The results indicate significant improvements in resource utilization and energy efficiency, positioning our framework as a sustainable and effective solution for zero-touch network and service management in 6G networks.