Using Reinforcement Learning and Game Theory for Determining Cooperative Nodes in Multi-hop Wireless Networks

Abstract

Multi-hop wireless networks, such as ad hoc and IoT networks, contain relay nodes offering packet relay between sources and destinations. Ensuring energy efficiency, minimizing delay, and maintaining reliable routing remain supreme challenges, especially when node cooperation is essential in guaranteeing network survivability. Unlike traditional approaches that assess cooperation globally and concerning the origin, this work proposes a new approach that detects cooperative nodes on a per-destination basis using a hybrid framework based on reinforcement learning and game theory. In the proposed framework, every node is an independent agent that learns optimal routing policies through ongoing interactions. Reinforcement learning allows dynamic adaptation, while game-theoretic modeling achieves incentive compatibility through payment for cooperative behavior and punishment for selfishness. Simulation results demonstrate that the algorithm presented outperforms benchmark algorithms including HChOA, QoS-RSIA, RL-based, and DQN-RSS. The suggested approach attains a 20% lessening in energy consumption, a 37.5% reduction in average delay, a 10.6% betterment in packet delivery ratio, a 14.1% improvement in cooperation ratio, and a 17.5% improvement in network lifetime. These enhancements validate the efficacy of the suggested approach in improving routing performance, reducing energy consumption, and guaranteeing reliability in dynamic multi-hop wireless networks.