Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/48C7C87F-11EF-4855-82FE-5E3C57F90EDD | ||
| Year | 2022 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Konstantinos Zacharopoulos, "Graph-Based modeling of cellular Hotspot data analysis", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2022 https://doi.org/10.26233/heallink.tuc.94674 | ||
| Appears in Collections |
Cellular networks have gone through significant changes in infrastructure during the last few decades. The technological advance along with the massive increase of population keep pushing the mobile network's capabilities to their limits. As the network usage requirements become larger, so do the radio expert's needs for accurate and timely information regarding the status of the cellular network. With that information at hand, they have the ability to foresee and prevent unwanted circumstances, such as network failure due to unmanageable overload. In this thesis, I propose a Neural Network structure which aims to make both swift and precise forecasts of such undesirable events. I make use of well-known Neural Network architectures such as the Graph Neural Network(GNN) and the Recurrent Neural Network(RNN), a combination which allows for monitoring and learning both spatial and temporal patterns that the cellular network may exhibit. In addition, a Graph partition is introduced, which effectively splits the original graph into much smaller and manageable sub-graphs with the idea of further increasing the Neural Networks performance metrics while also significantly scaling down its time complexity. Furthermore, I propose the addition of an Hierarchical model in the original architecture, an addition which nearly maximizes the precision in most use-cases. The proposed architecture succeeds in increasing the precision of its predictions compared to other known implementations. Moreover, it has a steady performance across various sizes of historical data provided to the network and across different targeted prediction horizons.
Copyright © DIAS 2013
