A Hybrid Genetic Reinforcement Learning Technique for Intelligent Cluster Head Selection in WSNs

Abstract

In Wireless Sensor Networks (WSNs), selection of Cluster Head (CH) in view of the optimum performance is a difficult task in view of constraints of energy resources, dynamic energy state of the network and dynamic data communication requirements, etc. Here, optimization of these can make better solutions. A Hybrid RL-GA Algorithm is developed to give the optimum solution for this, by combining a Genetic Algorithm (GA) with Q-learning. Optimization of both how CHs are selected and how the data is routed afterward was performed. Genetic Algorithm founds initial global structure, individual nodes and Cluster Heads take over using Q learning. This allows them to make adaptive routing decisions. Network continuously learns the most energy-efficient paths, and it recovers much faster from link failures. The GA provide a strong foundation, and Q-learning handles the day-to-day network stability. With simulations, our Hybrid RL-GA protocol outperformed the standalone RL method used in the latest research works. The total network lifetime is higher by roughly 50% than standalone RL.