Abstract: Optimization of sliding sleeve completion system involves complex problems such as multivariable, strongly coupled, and nonlinear characteristics. This study establishes a coupled flow productivity model for horizontal well sliding sleeve and reservoir, with the discrepancy between theoretical predictions and numerical simulation results validated to be within 12%. Based on this model, a reinforcement learning framework guided by Markov decision theory is adopted, and a PER-MADDPG algorithm is proposed for optimizing multi-stage sliding sleeve completion in horizontal wells by introducing a prioritized experience replay mechanism. Results show that compared with DDPG and MADDPG, the PER-MADDPG-optimized horizontal well sliding sleeve completion achieves the highest reward with minimal loss during training, increasing cumulative oil production by 10.7% and reducing comprehensive water cut by 6.4% after two years. By intelligently prioritizing and replaying experiences, this algorithm significantly enhances the learning efficiency and sample utilization of agents, providing theoretical guidance for real-time multi-agent control during oil well production.