Abstract: Intelligent reflective surfaces (IRSs) are recognized as an effective approach to enhance the efficiency of wireless power transfer (WPT). Many scholars have utilized a single IRS to assist wireless rechargeable sensor networks (WRSNs) in energy harvesting, further optimizing charging strategies to extend their lifespan. Compared with a single IRS, deploying multiple IRSs not only effectively prevents the loss of additional Beamforming (BF) gain caused by the failure of a single IRS but also supports the charging needs of multiple nodes simultaneously. However, the use of multiple IRSs leads to an "exponential" increase in the number of channels, thereby facing the "curse of dimensionality" problem when solving large- scale WRSNs charging strategies. To address this, a charging strategy for WRSNs assisted by multiple IRSs based on reinforcement learning is proposed to reduce computational complexity and resource consumption. A distributed method was adopted to optimize the phase shifts of multiple IRSs, aimed to maximize the receiving power of the nodes and shorten the solution time. The concept of game theory was introduced, transforming the single Markov decision process (MDP) model into a multi- MDP game model, thereby reducing the state space of the reinforcement learning model. The Nash Q- learning algorithm was employed to accelerate the search and solution process of the optimal charging strategy. Simulation results demonstrate that this method significantly enhances the lifespan and energy efficiency of WRSNs.