In large-scale dynamical Underwater Acoustic Sensor Networks (UASNs),node localization suffers from low node coverage,high cost,and substantial localization error.Here,we propose an iterative localization algorithm with Reinforcement Learning for Cumulative Error Correction (RL-CEC).First,we enhance the traditional Random Waypoint (RWP) mobility model by incorporating real ocean current data to adjust node movement speeds,thus better capturing the mobility patterns of underwater nodes.Second,we analyze the sources of cumulative errors during iterative localization and introduce a reinforcement learning algorithm to predict and correct these errors, thereby significantly improving localization accuracy.Simulation results show that RL-CEC outperforms several benchmark algorithms in both localization accuracy and time efficiency.Specifically,it reduces localization error by 22.7%,73.8%,40.7%,44.5%,17.8%,and 77.5% compared to LSVP,MBIL,MPL,MP-PSO,HNN,and LS,respectively.In terms of localization time,RL-CEC achieves reductions of 17.7%,30.5%,41.9%,and 35.3% compared to LSVP,MPL,MP-PSO,and HNN,respectively.Moreover,RL-CEC maintains moderate algorithmic complexity,strikes a favorable balance between accuracy and efficiency,and achieves a high node coverage rate.