Intelligent routing agent based on Q-learning and Markov decision processes for routing optimization in DTN networks

DTN networks present significant routing challenges due to their intermittent connectivity and node mobility. Traditional routing protocols, often pre-established, prove inadequate in adapting to the changing dynamics of DTN networks. To address these challenges, the integration of advanced artificial intelligence techniques, notably intelligent agents, is emerging as a promising solution. This article proposes an approach based on an intelligent agent optimized by Q-learning, capable of perceiving, analyzing, and acting autonomously to optimize routing in these dynamic environments. The intelligent agent integrates three main modules: a perception module to collect data in real time, a decision module based on Q-Learning to autonomously adapt its decisions and routing strategies, and an action module to execute the decisions made. The modeling of DTN networks as Markov Decision Processes (MDP) enables the agent to dynamically learn and select optimal routing strategies, without prior knowledge of network transitions. The simulations show that this approach improves routing performance compared to traditional protocols, with gains in terms of delivery rate, latency reduction, energy efficiency, and resource optimization. These results highlight the potential of intelligent agents to address the complex challenges of DTN networks. • This study proposes a Q-Learning-based intelligent agent to enhance routing performance in Delay Tolerant Networks (DTN). • Modeling of DTN environments as Markov Decision Processes (MDP) to formalize uncertainty and network dynamics. • Proposes a modular and adaptive architecture comprising three main modules: (i) Perception – for real-time network data collection, (ii) Decision – for routing optimization using Q-Learning, and (iii) Action – for dynamic execution of routing strategies.