Graph Neural Network Enhanced Dynamic Routing for UAV-Assisted Delay Tolerant Network
Ran Gao, Can Tan, Honglin Fang, Peng Yu · 2024
In certain extreme environments where infrastructure such as base stations is absent, connections between nodes can be intermittent. Traditional networks struggle to function in these conditions. Delay Tolerant Network (DTN) are a network architecture designed for such scenarios. Utilizing Unmanned Aerial Vehicle (UAV) as nodes within DTN can enhance the network’s transmission capabilities and compensate for the deficiencies of ground nodes. However, previous routing methods have shortcomings. Given the high speed and greater freedom of movement of UAVs compared to ground nodes, the network topology is subject to constant and significant change. Traditional routing methods, which typically rely on predetermined rules for selecting forwarding nodes, struggle to adapt to such variations. This paper introduces Graph Neural Network (GNN) into routing decisions for DTN to learn network structural information and assist with routing decisions. In this paper, we first model the routing problem within DTN, accounting for the dynamic changes in network topology and limited resources. Subsequently, we propose a GNN-based dynamic routing method. This method utilizes a graph neural network model that takes as input the feature s of nodes at the current moment and the network’s connectivity information to generate optimal forwarding nodes for each data packet. The proposed method has been implemented in the Opportunistic Networking Environment (ONE) simulator. Experimental results indicate that our method achieves higher delivery ratio, lower latency, and reduced overhead compared to traditional methods.