Dynamic charging location determination for energy level equalization optimization in wireless rechargeable sensor networks

In wireless rechargeable sensor networks, broadcast charging provides a promising and efficient node energy replenishment approach that allows mobile chargers (MCs) to charge multiple nodes simultaneously. However, MCs usually charge nodes at a fixed location in existing work, which may result in large variability in residual energy levels after charging. Such non-equalization in node energy levels can cause degradation of network robustness and quality of service, shortened network lifetime, and increased maintenance cost. Therefore, this paper focuses on optimizing strategy in broadcast charging to achieve fine energy redistribution, thus maximizing the equalization of node energy levels. Specifically, the energy difference index is first defined to characterize the equalization of node energy levels. Second, domain discretization is utilized to characterize the MC-accessible charging locations. Finally, a novel dynamic charging location determination algorithm (DCLD) based on reinforcement learning (RL) is proposed, which takes into account the dynamic changes of MC charging locations and realizes a fine-grained node received energy redistribution to improve the equalization of node energy level. In DCLD, RL is introduced to explore MC discrete candidate charging locations autonomously, and the dual Q-table update mechanism reduces the overestimation error. In addition, a prioritized replay buffer mechanism is applied to filter and utilize the high-value experiences to make the learning process faster and more stable. Extensive simulations show that DCLD significantly outperforms other approaches.