A Lifetime-Aware Centralized Routing Protocol for Wireless Sensor Networks using Reinforcement Learning

This paper presents the design of a Lifetime-Aware Centralized Q-routing Protocol (LACQRP) for Wireless Sensor Network (WSN) to maximize the network lifetime. This is achieved by implementing Q-learning on the sink of the WSN, which also acts as a controller that has global knowledge of the network topology as enabled by Software-Defined WSN (SDWSN). The controller generates all possible distance-based minimum spanning trees (MSTs), which form the set of routing tables (RTs). The maximization of the network lifetime is achieved by the controller learning the routing table that minimizes the maximum of the sensor nodes' consumption energies using Reinforcement Learning (RL). The simulation results show that the LACQRP learns the best RT that maximizes the network lifetime and has a better network lifetime performance when compared with recent distributed RL routing protocols for lifetime optimization, which are Reinforcement Learning-Based Routing (RLBR) and Reinforcement Learning for Lifetime Optimization (R2LTO).