Relay Nodes Selection Using Reinforcement Learning

In IoT networks, the nodes work cooperatively. They receive data packets and re-transmit to the sink node (or fusion node) via multiple relay nodes. In order to reduce the loss of packets as well as power consumption, it is important to transmit data packet successfully and find an optimal path from source node to sink node. Relay node selection is one of key research challenges in IoT networks. The reinforcement learning (RL) deals with sequential decision making problem under uncertainty. The goal of sequential decision making problem is to select actions to maximize long term rewards. The RL has emerged as a powerful method for many different areas. In this paper, relay node selection problem in IoT networks with channel measurement data is formulated as a Markov decision process (MDP) problem. The relay node selection problem is solved using Q learning when a local channel measurement map is given. We find an optimal relay node selection path.