基于注意力机制和价值强化学习的WRSN一对多充电调度方法

Attention Mechanism and Value Reinforcement Learning Based One-to-Multiple Charging Scheduling Scheme In WRSN

在大规模无线可充电传感器网络(WRSN)中,一对一充电模型难以满足节点巨大能量需求,充电效率更高的一对多充电成为更合理选择。提出了一种基于注意力机制和价值强化学习的WRSN在线一对多充电调度方法(MAQRL),从充电序列和充电时长两方面优化移动充电设备(MC)调度。首先,基于MC有效充电范围覆盖最多节点对网络内节点进行分簇处理,并基于价值强化学习优化充电序列。MAQRL结合注意力机制和价值强化学习,利用注意力机制提取特征和MC对节点的注意力,利用双价值强化学习来减少高估,以提高充电方法的充电性能;其次,通过分析整个网络中节点的平均剩余生存时长和MC平均移动延迟,动态优化充电时间,减少后续节点因等待时间过长而导致的死亡。大量的仿真实验表明,与现有几种充电方法相比,MAQRL在降低节点死亡率和充电延迟方面具有显著优势。

In large-scale wireless rechargeable sensor networks(WRSNs),the one-to-one charging mode can hardly meet the huge energy demand of nodes,and the one-to-multiple charging mode with higher charging efficiency becomes a more reasonable choice.An online one-to-multiple charging scheduling scheme(MAQRL)for WRSNs based on attention mechanism and value learning is proposed to optimize the mobile charging device(MC)scheduling in terms of charging sequence and charging time.Firstly,the nodes in the net-work are clustered based on the MC effective charging range covering the most nodes,and the charging sequence is optimized based on value reinforcement learning.By combining attention mechanism and value reinforcement learning,MAQRL uses attention mechanism to extract features and MC’s attention to nodes,and uses double value reinforcement learning to reduce overestimation,to improve the char-ging performance of the charging scheme.Secondly,by analyzing the average remaining living time of nodes in the whole network and the average movement delay of MC,the charging time is dynamically optimized to reduce the death of subsequent nodes caused by too long waiting time.Extensive simulation experiments show that MAQRL has better performance in reducing node mortality and charging delay compared with several existing charging schemes.