AI-Driven Optimization of CloudRAN Test Platforms through Digital Twin Technology

Sammanfattning

This thesis investigates the optimization of Cloud Radio Access Network (C-RAN) test platform utilization through the application of Artificial Intelligence (AI) and Digital Twin technology, with the goal of reducing operational costs and environmental impact. A predictive framework was developed to forecast test channel usage by analyzing time series data radio bearer (DRB) and radio resource control (RRC) metrics collected from an Ericsson C-RAN test environment using Prometheus. Utilization metrics based on DRB traffic were defined in collaboration with Subject Matter Experts (SMEs), and several AI forecasting models were evaluated. The LightGBM model, optimized through SME-informed DRB thresholds (150 MB) demonstrated the best performance with an F1-score of 0.86, precision of 0.85, and recall of 0.87. The most predictive features were recent DRB traffic patterns and time related indicators such as hour of day. A conceptual Digital Twin interface was prototyped in Figma, following Ericsson Design System (EDS) guidelines, to visualize predicted channel availability and support scheduling decisions. The study concludes that AI-driven prediction of C-RAN test channel utilization is a viable approach to improving resource management. Integrating AI models with Digital Twin visualization offers a practical solution for enhancing efficiency, reducing costs, and promoting sustainable telecom testing. Future work includes real-time integration and further refinement of the predictive models.