Methodology for forecasting user mobility in 5G networks

DOI: 10.31673/2412-4338.2024.041861

Abstract

The article is devoted to the development of a methodology for predicting user mobility in 5G networks using machine learning, spatiotemporal correlation and mathematical modeling. The main goal of the work is to increase the accuracy of predicting user movements, taking into account the specifics of 5G networks, such as high data rates, limited base station resources and complex network topology. The article presents a detailed formalization of the problem, where the user's mobility is described through a discrete set of states corresponding to its position in space and time. An objective function is proposed to minimize the discrepancy between the predicted and real probabilities of user movement. The spatiotemporal model takes into account physical, topological and network constraints, such as movement speed, availability of transitions between states and base station capacity. The methodology is based on the use of recurrent neural networks (RNN), which allow modeling dependencies in time and space. The model training process includes parameter optimization using stochastic gradient descent with regularization, which improves the algorithm's convergence. Metrics such as Accuracy, F1-score, and Log-loss are used to assess the prediction accuracy. The research results confirm the effectiveness of the method in scenarios with high user movement speed and complex network structure. Integration of the model into the network management system allows for dynamic optimization of resource allocation, reduction of congestion probability, and improvement of service quality. Future research prospects include adaptation of the method for next-generation networks and integration of new data sources, such as information from satellite systems.

Keywords: 5G network, information system, mobility, prediction methods, machine learning methods, neural networks, information technology.

References

1. Melnyk, O. A. Development of user mobility prediction models for adaptive communication systems. Computer Technologies and Systems. 2022. No. 11. P. 15–29.
2. Data-driven 5G handover optimization: a comparative analysis of machine learning techniques / P. Mehta et al. Wireless Communications and Mobile Computing. 2023. Vol. 2023. P. 8854776.
3. Predicting mobile user location using hybrid models of machine learning and Markov chains / D. Chen et al. Journal of Computational Science. 2021. Vol. 56. P. 101409.
4. Deep reinforcement learning for mobility management in ultra-dense 5G networks / A. Wang et al. IEEE Internet of Things Journal. 2022. Vol. 9, no. 3. P. 1765–1778. URL: https://doi.org/10.1109/JIOT.2021.3055643.
5. Traffic and mobility prediction for network optimization in 5G / L. Kumar et al. ACM Transactions on Internet Technology. 2020. Vol. 21, no. 2. P. 1–20.
6. Ivanenko, M. S. Models of user behavior analysis in mobile networks: prediction and clustering. Ukrainian Journal of Information Technologies. 2023. No. 13, issue 2. P. 45–60.
7. Advanced AI-based mobility management for next-generation networks / F. Costa et al. IEEE Network. 2021. Vol. 35, no. 4. P. 36–43. URL: https://doi.org/10.1109/MNET.2021.3054017.
8. User behavior modeling for 5G: challenges and approaches / K. Singh et al. Wireless Personal Communications. 2021. Vol. 119, no. 3. P. 2111–2130.
9. Machine learning models for spatio-temporal prediction in mobile networks / P. Wu et al. IEEE Communications Surveys & Tutorials. 2020. Vol. 22, no. 4. P. 2452–2470.
10. Boyko, T. G. Using deep learning to predict mobility in large networks. Bulletin of NTUU "KPI". Series: Computer Sciences. 2022. No. 29. P. 11–22.