HGC-LSTM: A Graph Neural Network-based Model for HO Forecasting in Mobile Networks
Résumé
In the realm of mobile networks, the escalating growth in data traffic, primarily fueled by the proliferation of connected devices and the rising demand for data-intensive applications, poses an ongoing challenge. Ensuring seamless connectivity, mobility, and optimal user experience in such a dynamic environment is more complex than ever. In this context, effective user mobility management through handovers (HOs) emerges as a critical task. Well-managed HOs contribute to enhanced user quality of service (QoS), minimizing disruptions in connectivity as users move within the network. Conversely, poor HO management can lead to issues such as increased latency, network congestion, and elevated operational costs. Against this backdrop, our work focuses on the problem of HO forecasting, aiming to predict HOs among base stations over time. To address this challenge, we introduce the HO Graph Convolutional Long Short-Term Memory (HGC-LSTM) neural network forecasting approach. This innovative methodology incorporates Graph Neural Networks (GNNs) and Long Short-Term Memory (LSTM) networks to capture spatio-temporal dependencies and correlations among neighboring pairs of base stations during the forecasting process. Our evaluation, conducted on a real-world dataset, demonstrates that the proposed HGC-LSTM approach surpasses state-of-the-art methods in the likes of MuLSTM, transformer and ARIMA, reaching the desired trade-off between user QoS and overprovisioned resources.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |