| dc.description.abstract | This research addresses three critical questions in the context of enhancing 5G network coverage via UAV-assisted communication: predicting user mobility, optimizing UAV deployment, and measuring the performance of developed algorithms. The primary goal is to utilize machine learning (ML) models, specifically Long Short-Term Memory (LSTM) networks and other neural networks (NNs), to predict the geographical location of users at given times. This capability enables more effective resource management and dynamic Unmanned Aerial Vehicle (UAV) placement in 5G networks.
Predicting user mobility involves analyzing historical movement data using ML algorithms. These models learn from past behaviors to forecast future user locations, enhancing resource management and UAV deployment efficiency. Optimal UAV deployment aims to position UAVs strategically to maximize coverage and ensure reliable communication in a 5G network. Factors such as user location, UAV transmit power, and network requirements are considered to determine the best UAV positions. The performance of these ML algorithms are measured by criteria such as accuracy, computational speed, coverage, and generalizability across different scenarios.
The research methodology involves the use of the Geolife dataset, comprised of extensive Global Positioning System (GPS) trajectory data from 182 users over five years, predominantly in Beijing. This dataset enables the ML models to learn and predict user mobility patterns. For optimal UAV deployment, various schemes were evaluated, including equal horizontal sections, random selection, and calculated distance selection schemes. The coverage area for UAVs was calculated and visualized, demonstrating the effectiveness of different deployment strategies. The implementation was carried out in Python using Jupyter Notebooks, where the dataset was cleaned, standardized, and split into training and testing sets before being applied to the ML models. The training results showed significant improvements in user mobility prediction, with most models achieving over 80\% accuracy. The K-Nearest Neighbors (KNN) model performed exceptionally well, with accuracy ranging from 96\% to 99\%.
UAV deployment strategies were analyzed through simulations, showing that calculated distance selection schemes provided the best coverage with minimal overlap. The study also investigated the impact of increasing alpha and height values on the number of users served within the UAV coverage area, finding significant improvements with increased values. This research demonstrates the feasibility and effectiveness of using ML models for predicting user mobility and optimizing UAV deployment in 5G networks. The results highlight the potential for improved network coverage and resource allocation, paving the way for more dynamic and efficient 5G communication systems. | |