Ammonia production using renewable energy holds significant potential as a zero-carbon fuel carrier in energy systems. This paper proposes an energy management model based on tiered carbon trading for multi-energy microgrid that incorporate re-newable energy-based ammonia production. First, a full-system model was developed that encompasses photovoltaic power generation, electrolytic hydrogen production, hydrogen storage, ammonia synthesis, ammonia storage, and combined heat and power (CHP), providing a detailed description of the ther-mo-electrochemical dynamics of the ammonia synthesis process. Second, to address fluctuations in renewable energy output, the energy management problem is modeled as a Markov decision process (MDP) with unknown state transition functions. Fur-thermore, based on a “centralized training–decentralized execu-tion” multi-agent deep reinforcement learning (MADRL) archi-tecture, a model-free, data-driven optimization algorithm is pro-posed, and the agent learning mechanism is designed using a soft actor-critic (SAC) framework. Case studies demonstrate that the proposed model can achieve low-carbon economic dispatch for multi-microgrids under fluctuating renewable energy conditions. The proposed algorithm outperforms traditional optimization methods in both solution speed and accuracy.