ML-based 5G Core Network Load Forecasting with Metrics from Performance Management

As the number of connected devices continues to increase and the complexity of services supported by 5G grows, significant challenges may arise in managing the important signaling and processing of the control plane. This will result in extensive resource utilization. While deploying additional resources may be relatively straightforward, the decision-making process for resource scaling poses a challenge. Hence, this paper provides the 5G network management system with the capability of system load prediction to assist in the decision-making process of resource scaling. To apply system load prediction to the 5G core (5GC) network, we collect real-time performance measurement data from the 5GC network using open-source 5G software. We utilize this data to train neural networks to predict the load of 5GC network functions. Additionally, we design a specific loss function for the prediction model to align with our resource scaling strategy objectives. Experimental results demonstrate that the application of performance measurements enhances the accuracy of load prediction tasks, thereby contributing to more effective resource management.