Multi-Agent Reinforcement Learning for Carbon Neutrality in Urban Energy Grids

The integration of intermittent renewable energy sources (RES) into aging urban power grids presents a significant barrier to achieving 2030 carbon neutrality goals. Traditional supervisory control and data acquisition (SCADA) systems are increasingly unable to manage the bidirectional energy flows introduced by residential solar arrays and electric vehicle (EV) charging stations. This paper proposes a decentralized control architecture using Multi-Agent Reinforcement Learning (MARL) to optimize grid stability and minimize carbon intensity. We introduce the "Nexus-Alpha" algorithm, which empowers local substations to operate as autonomous agents that negotiate energy distribution based on real-time carbon pricing and demand forecasting. Using a high-fidelity simulation calibrated with PJM Interconnection utility data from 2025, our model achieved a 14.2% reduction in peak-load emissions and an 8.5% improvement in voltage regulation. This research provides a scalable framework for transitioning to self-organizing smart grids