Emami, Negar (2023). Deep Learning Techniques for Mobility Prediction and Management in Mobile Networks. (Thesis). Universität Bern, Bern
|
Text
23emami_n.pdf - Thesis Available under License Creative Commons: Attribution-Noncommercial-No Derivative Works (CC-BY-NC-ND 4.0). Download (14MB) | Preview |
Abstract
Trajectory prediction is an important research topic in modern mobile networks (e.g., 5G and beyond 5G) to enhance the network quality of service by accurately predicting the future locations of mobile users, such as pedestrians and vehicles, based on their past mobility patterns. A trajectory is defined as the sequence of locations the user visits over time. The primary objective of this thesis is to improve the modeling of mobility data and establish personalized, scalable, collective-intelligent, distributed, and strategic trajectory prediction techniques that can effectively adapt to the dynamics of urban environments in order to facilitate the optimal delivery of mobility-aware network services. Our proposed approaches aim to increase the accuracy of trajectory prediction while minimizing communication and computational costs leading to more efficient mobile networks. The thesis begins by introducing a personalized trajectory prediction technique using deep learning and reinforcement learning. It adapts the neural network architecture to capture the distinct characteristics of mobile users’ data. Furthermore, it introduces advanced anticipatory handover management and dynamic service migration techniques that optimize network management using our high-performance trajectory predictor. This approach ensures seamless connectivity and proactively migrates network services, enhancing the quality of service in dense wireless networks. The second contribution of the thesis introduces cluster-level prediction to extend the reinforcement learning-based trajectory prediction, addressing scalability challenges in large-scale networks. Cluster-level trajectory prediction leverages users’ similarities within clusters to train only a few representatives. This enables efficient transfer learning of pre-trained mobility models and reduces computational overhead enhancing the network scalability. The third contribution proposes a collaborative social-aware multi-agent trajectory prediction technique that accounts for the interactions between multiple intra-cluster agents in a dynamic urban environment, increasing the prediction accuracy but decreasing the algorithm complexity and computational resource usage. The fourth contribution proposes a federated learning-driven multi-agent trajectory prediction technique that leverages the collaborative power of multiple local data sources in a decentralized manner to enhance user privacy and improve the accuracy of trajectory prediction while jointly minimizing computational and communication costs. The fifth contribution proposes a game theoretic non-cooperative multi-agent prediction technique that considers the strategic behaviors among competitive inter-cluster mobile users. The proposed approaches are evaluated on small-scale and large-scale location-based mobility datasets, where locations could be GPS coordinates or cellular base station IDs. Our experiments demonstrate that our proposed approaches outperform state-of-the-art trajectory prediction methods making significant contributions to the field of mobile networks.
| Item Type: | Thesis |
|---|---|
| Dissertation Type: | Single |
| Date of Defense: | 27 September 2023 |
| Subjects: | 000 Computer science, knowledge & systems 500 Science > 510 Mathematics |
| Institute / Center: | 08 Faculty of Science > Institute of Computer Science (INF) > Communication and Distributed Systems (CDS) |
| Depositing User: | Sarah Stalder |
| Date Deposited: | 06 Nov 2023 17:25 |
| Last Modified: | 27 Sep 2024 22:25 |
| URI: | https://boristheses.unibe.ch/id/eprint/4693 |
Actions (login required)
 |
View Item |