Low-carbon schedule of electricity user based on consensus-based multi-agent proximal policy optimization

With the intensification of climate change, the power industry urgently requires low-carbon transformation, yet traditional centralized dispatch methods face challenges in balancing user privacy protection and efficient resource coordination. While distributed consensus-based approaches and reinforcement learning (RL) have shown promise in addressing these issues, existing RL methods often neglect privacy considerations or suffer from high training complexity in stochastic power systems. This study proposes a Consensus-based Multi-Agent Proximal Policy Optimization (CMA-PPO) framework for low-carbon emission load dispatch. In this framework, distributed agents deployed at power user nodes collaboratively optimize dispatch strategies through localized training and consensus-driven coordination.