Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Electrical-Electronic Engineering

This thesis introduces a method to enhance the stability and efficiency of vehicular clusters in Vehicular Ad Hoc Networks (VANETs) by utilizing a clustering algorithm based on Graph Neural Networks (GNNs). As the number of vehicles on the road increases, problems such as traffic congestion, energy inefficiency, and air pollution have become more severe. This work tackles these issues by improving the stability of vehicle clusters, thereby boosting the efficiency of cooperative driving. Unlike conventional techniques that rely on periodic communication and the selection of cluster heads (CH), this approach uses a GNN model to create effective node representations, grouping vehicles with similar behaviours into stable clusters. The performance of the clustering methods was rigorously assessed using the open source highD dataset. The results demonstrated superior cluster longevity and efficiency compared to the K-means algorithm. The GNN model adeptly processes vehicular features, including speed, position, and acceleration, alongside graph data, using a force-directed algorithm to compute vehicle connectivity metrics. This innovative approach significantly reduces the overhead of control messages, thereby enhancing the overall system stability. The results of this research demonstrate that the GNN-based clustering algorithm, which incorporates both vehicular characteristics and graph structures, significantly surpasses traditional clustering methods. This positions it as a highly promising solution for advancing future intelligent transportation systems. By improving the management of vehicular clusters, this method contributes to more efficient and sustainable transportation networks, potentially leading to reduced traffic congestion, lower energy consumption, and diminished air pollution.