Deep Reinforcement Learning-Based Dynamic Charging-Recycling Scheme for Wireless Rechargeable Sensor Networks

To meet the huge energy demand of large-scale wireless rechargeable sensor networks (WRSNs), using multiple wireless charging vehicles (MCVs) is a natural way, however, which also results in high costs of construction and operation. A new separable charging mode is regarded as a feasible and low-cost method to replenish energy for large-scale WRSN. Nonetheless, existing works consider the deployment and recycling of wireless chargers as two completely separate phases neglect the intrinsic interactions between them, and thus still suffer performance problem. To improve charging performance, we study the dynamic charging-recycling scheduling (DCRS) problem, and prove that it is NP-hard. Then, we propose a deep reinforcement learning (DRL)-based scheme to get a near optimal solution of the DCRS problem. The proposed scheme utilizes double deep Q-learning network (DDQN) to jointly optimize the charging and recycling scheduling of MCV, responding to more charging requests through timely recycling wireless chargers. The Q-network extracts the feature information of WRSN and determines whether and when to charge a sensor node or recycle a wireless charger. The target Q-network optimizes the charging-recycling strategy, which can alleviate the overestimation problem and improve the stability of training. Furthermore, we analyze the minimum number of wireless chargers under the condition of ensuring a WRSN run stably. The simulation results show that the proposed scheme outperforms the state-of-the-art scheme in terms of number of dead nodes and charging delay.