5G Core Network Traffic Prediction Based on NWDAF Multi-Model Fusion

Large-scale device access in 5G may exceed core network resource constraints, affecting user demand. Analyzing and predicting core network traffic is vital to optimise network configuration, improve quality of service and reduce costs. However, accurately predicting network traffic while ensuring data privacy is a challenging task due to the diversity, complexity, and sensitivity of core network traffic data. Inspired by Network Data Analytics Function(NWDAF) and ensemble learning, we propose NWDAF-MMF, a multi-model fusion network traffic prediction framework based on NWDAF. In NWDAF-MMF, NWDAF obtains data from Network Functions(NFs), reducing the need for numerous data transmissions and storage. Additionally, multi-model fusion can process different types of data provided by NFs. Our method comprises three submodules: the data preprocessing module, the multi-model training and analysis module, and the multi-model fusion module. The data processing module is responsible for data aggregation, data cleaning, feature selection, and dataset division provided by NFs. The multi-model training and analysis module is responsible for training the base model and obtaining its prediction results. The multi-model fusion module is responsible for integrating, optimizing, and outputting the final prediction results. We evaluate the proposed method using real network datasets. The experimental results show improved performance in MAE, RSME, and MAPE by 12.32%, 6.40%, and 12.66% compared to the single model, respectively.