RL-Sleep: Temperature Adaptive Sleep Scheduling using Reinforcement Learning for Sustainable Connectivity in Wireless Sensor Networks

• Published in: Sustainable computing informatics and systems Vol. 26; p. 100380
• Main Authors: Banerjee, Partha Sarathi, Mandal, Satyendra Nath, De, Debashis, Maiti, Biswajit
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.06.2020
• Subjects: Entropy; Heterogeneous wireless sensor network; Intelligent sleep scheduling; Reinforcement learning; Temperature; Temperature; Intelligent sleep scheduling; Entropy; Heterogeneous wireless sensor network; Reinforcement learning
• ISSN: 2210-5379
• DOI: 10.1016/j.suscom.2020.100380

[Display omitted] •RL-SLEEP protects sensor nodes from faster energy-depletion caused by temperature.•RL-SLEEP enables the sensor nodes to perform state transition depending on atmospheric temperature variation.•RL-SLEEP defines intermediate states between Active and Sleep states and the sensor nodes can intelligently switch between states.•RL-SLEEP gives due priority to the status of neighborhood before switching to a new state and thus provides better sustainability for the network.•RL-SLEEP performs better than BMAC and SOPC with respect to packet-delivery-ratio, average lifetime of nodes, and probability of connectivity. Temperature variations have a significant effect on the sustainable operation of the power-constrained wireless sensor networks. The characteristics of wireless communication links deteriorates considerably with increase of temperature. Proactive measures may not always perform well in a dynamic environment where both the wireless links and sensor nodes are supposed to behave unexpectedly. Environment adaptive efficient sleep-schedule strategy can preserve the resources of the low power sensor nodes and thereby alleviate the adverse effects of temperature. In this paper, temperature adaptive intelligent sleep-scheduling strategy (RL-Sleep) for the wireless sensor nodes has been proposed. This algorithm is based on Reinforcement Learning which enables a node in the network to perceive the environment and decide autonomously about the action (transmit, listen or sleep) conducive for a stable operation of the network. Simulation results exhibit a good performance of the proposed approach in terms of sustainable operations of the network and connectivity.