Multi-Decision Dynamic Intelligent Routing Protocol for Delay-Tolerant Networks

Delay-tolerant networks face challenges in efficiently utilizing network resources and real-time sensing of node and message statuses due to the dynamic changes in their topology. In this paper, we propose a Multi-Decision Dynamic Intelligent (MDDI) routing protocol based on double Q-learning, node relationships, and message attributes to achieve efficient message transmission. In the proposed protocol, the entire network is considered a reinforcement learning environment, with all mobile nodes treated as intelligent agents. Each node maintains two Q-tables, which store the Q-values corresponding to when a node forwards a message to a neighboring node. These Q-values are also related to the network’s average latency and average hop count. Additionally, we introduce node relationships to further optimize route selection. Nodes are categorized into three types of relationships: friends, colleagues, and strangers, based on historical interaction information, and message forwarding counts and remaining time are incorporated into the decision-making process. This protocol comprehensively takes into account the attributes of various resources in the network, enabling the dynamic adjustment of message-forwarding decisions as the network evolves. Simulation results show that the proposed multi-decision dynamic intelligent routing protocol achieves the highest message delivery rate as well as the lowest latency and overhead in all states of the network compared with other related routing protocols for DTNs.