To address the multi-ship collision avoidance problem in the context of growing onboard intelligence, a cooperative collision avoidance game model based on multi-agent reinforcement learning was developed using the route exchange mechanism. Real-time sharing and negotiation of intended route information among ships were enabled. The multi-ship collision avoidance decision was transformed into a multi-agent cooperative game model, with each ship possessing independent decision-making and execution capabilities and being driven by rationality and economic considerations. The objective is to optimize navigational efficiency, minimize collision risk, and comply with anti-collision rules. The multi-agent deep deterministic policy gradient algorithm was employed within a centralized training with decentralized execution framework to optimize collision avoidance strategies, enabling an approach to the Pareto optimal solution. Simulation results demonstrate that optimized routes obtained through reasonable heading adjustments effectively avoid collision zones, balancing safety, compliance, and navigational efficiency. The model that integrates multi-agent reinforcement learning and game theory provides a feasible solution for intelligent ship collision avoidance decisions under the E-navigation paradigm.