Transformer-Based Wireless Traffic Prediction and Network Optimization in O-RAN

Abstract

This paper introduces an innovative method for predicting wireless network traffic in concise temporal intervals for Open Radio Access Networks (O-RAN) using a transformer architecture, which is the machine learning model behind generative AI tools. Depending on the anticipated traffic, the system either launches a reinforcement learning-based traffic steering xApp or a cell sleeping rApp to enhance performance metrics like throughput or energy efficiency. Our simulation results demonstrate that the proposed traffic prediction-based network optimization mechanism matches the performance of standalone RAN applications (rApps/ xApps) that are always on during the whole simulation time while offering on-demand activation. This feature is particularly advantageous during instances of abrupt fluctuations in traffic volume. Rather than persistently operating specific applications irrespective of the actual incoming traffic conditions, the proposed prediction-based method increases the average energy efficiency by 39.7% compared to the "Always on Traffic Steering xApp" and achieves 10.1% increase in throughput compared to the "Always on Cell Sleeping rApp". The simulation has been conducted over 24 hours, emulating a whole day traffic pattern for a dense urban area.