Reinforcement learning based mobile charging sequence scheduling algorithm for optimal stochastic event detection in wireless rechargeable sensor networks

Abstract

Mobile charging provides a new way for energy replenishment in Wireless Rechargeable Sensor Network (WRSN), where the Mobile Charger (MC) is employed for charging sensor nodes sequentially according to the mobile charging sequence scheduling result. Event detection is an essential application of WRSN, but when the events occur stochastically, Mobile Charging Sequence Scheduling for Optimal Stochastic Event Detection (MCSS-OSED) is difficult and challenging, and the non-deterministic detection property of the sensor makes MCSS-OSED complicated further. This paper proposes a novel Multistage Exploration Q-learning Algorithm (MEQA) for MCSS-OSED based on reinforcement learning. In MEQA, MC is taken as the agent to explore the space of the mobile charging sequences via the interactions with the environment for the optimal Quality of Event Detection (QED) evaluated by both considering the sensing probability of the sensor and the probability that events may occur in the monitoring region. Particularly, a new multistage exploration policy is designed for MC to improve the exploration efficiency by selecting the current suboptimal actions with a certain probability, and a novel reward function is presented to evaluate the MC charging action according to the real-time detection contribution of the sensor. Simulation results show that MEQA is efficient for MCSS-OSED and superior to the existing classical algorithms.