AI-Driven Telecom System Optimization and 5G Network Management: Transforming Manufacturing and Enhancing Connectivity for Medical Devices and Financial Services

Abstract

In this paper, a novel use case of the AI-driven self-organizing framework for 5G networks is presented, which is an application of a machine learning-based real-time telecom resources management. The problem considered in this work is the potential use of a federated learning approach for telecom self-organizing deep reinforcement learning systems, particularly for real resources management in online mode. Three options for possible AI-driven network management based on federated learning are proposed: 1) a supervised machine learning model, trained on a telemetry database in the cloud, with model updates and downloads from the cloud to all base stations; 2) a distributed Q-learning model (network self-tuning) via a federated learning algorithm; 3) a reinforcement learning problem with shared experience with a local neural network. In the proposed approaches, advanced deep neural networks are capable of solving complex tasks, superior deep Q-learning combinations to current operations and maintenance layer systems and a significant improvement in the key performance indicator system of telecom operators.