Intelligent Scheduling in Wireless Sensor Networks using Reinforcement Learning based Deep Q-Networks

The basic challenge in spatially distributed resource restricted Wireless Senor Network (WSN) is to prolong the network lifetime maintaining its performance criterions like energy efficiency, coverage and connectivity. To achieve this, scheduling can be performed in sensor nodes in terms of their wake up time periods. There are many traditional scheduling mechanisms developed. But, sensor node scheduling still poses a challenge when the WSN becomes dynamic in nature. In this paper, Deep Q-network modeling is carried out with Reinforcement scheduler. The advancement in reinforcement learning and particularly Deep Q-Network (DQN) algorithms provides an efficient tool for implementing intelligent scheduling. When an agent is trained using these techniques, its ability to differentiate between events improves thus, optimizing the scheduling process. The presented scheduling algorithm is evaluated and compared with traditional scheduling algorithms for the same network characteristics. The DQN modeling shows promising potential to intelligently schedule the sensor nodes in terms of energy consumption, average rewards and also minimum and maximum latencies.