Reinforcement Learning Based Congestion Control Mechanism for Opportunistic Networks

This work proposes a Reinforcement Learning-based Congestion Control protocol (called (RLCC)) for Opportunistic Networks to find the optimal number of message counts based on the real-time density in the network. (RLCC) jointly uses Q-learning and fuzzy logic to make the routing decision based on real attributes such as social status, centrality, activeness, message lifetime, hop count, and battery status. In the proposed (RLCC) scheme, the message priority is required to maintain the optimal count of messages in the network, and the fuzzy inference rules perform well in predicting the best hop for message transmission. All network nodes get inputs from the environment and take action accordingly. If the message is transferred successfully to the intended node, then the node receives the reward for the action, otherwise, the penalty will be assigned. Based on this, network nodes only select those nodes, which are capable to transmit the message from one node to another node. Simulation results demonstrate that RLCC is superior to the MARLCC and F-GSAF routing protocols using the infocom2006 real mobility data trace, in terms of delivery probability, latency, and overhead ratio.