A reinforcement learning based routing protocol for software-defined networking enabled wireless sensor network forest fire detection

Abstract

Critical event reporting Wireless Sensor Networks (WSNs) applications need vital requirements (extended network lifetime, reliability, real time responsiveness, and scalability) to be met to ensure outstanding efficiency. Previous frameworks only consider few individual requirements, thus ignoring the other equally important ones. Ensuring that an active path is available at all times is crucial for enabling the timely transmission of critical data and maintaining the quality of service required to efficiently support delay-sensitive applications. This paper proposes an application-specific Routing Protocol based on Reinforcement Learning (RL) for Software Defined Network (SDN)-enabled WSN forest fire detection (RPLS). First, we designed a clustering algorithm that delays re-clustering to save energy by keeping the same topology for several rounds. Unlike existing works, this algorithm decreases the cluster radius based not only on the energy parameters but also on the quality of the links. After the network clustering, the power of the SDN controller is used to intelligently define using RL the optimal paths for the sensor nodes and accordingly reduce the load on these constrained nodes. For routing strategy, we formulate an RL-based reward function considering not only the energy efficiency parameters but also the anticipated and post-failure reliability parameters to ensure real time responsiveness and optimize energy consumption. Finally, we conducted comparisons by means of simulations in forest fires detection scenario. Compared to RL-SDWSN, the results show an improvement of 14.064 % in network operational lifetime and 16.41 % in response time.