基于强化学习的无线传感器网络充电策略研究

Research on Charging Strategy of Wireless Sensor Network Based on Reinforcement Learning

基于移动充电器(MC)的充电策略一直是无线可充电传感器网络(WRSN)研究的热点。现有的充电策略通常是根据局部的网络信息实时选择充电节点,由于缺少全局信息无法保证充电策略的公平性,导致部分节点无法及时得到充电。针对此问题,设计了一种新颖的按需充电策略,首先,考虑节点能耗率的显著差异,给出了节点剩余寿命的计算方法。并开发一个双Q学习框架来优化充电策略,命名为Q-Charge充电策略,为了加快此充电策略中智能体的学习速度,在智能体学习策略中引入启发式学习策略,即有目的性的动作选择取代随机动作选择。仿真结果表明,Q-Charge充电策略不仅提高了充电效率,而且加快了MC移动路径的收敛速度。

The charging strategy based on Mobile Charger(MC)has always been the research hotspot of Wireless Rechargeable Sensor Networks(WRSN).The existing charging strategy generally makes the selection of charging nodes in real-time according to the local network information.The lack of global information makes the fairness of the charging strategy unable to be guaranteed,resulting in some nodes being unable to be charged in time.To solve this problem,a novel on-demand charging strategy is designed.Firstly,considering the significant difference in the energy consumption rate of nodes,a calculation method for the remaining life of nodes is proposed.A dual-Q learning framework is developed to optimize the charging strategy,named Q-Charge charging strategy.To speed up the learning speed of agents in this charging strategy,a heuristic learning strategy is introduced into the agent learning strategy to replace random action selection with purposeful action selection.The simulation results show that the Q-Charge charging strategy not only improves the charging efficiency but also speeds up the convergence speed of the MC moving path.