The decarbonization of modern energy systems requires intelligent control strategies capable of coordinating diverse and interconnected energy carriers. Multi-Energy Systems (MES), which integrate electricity, heat, gas, hydrogen, and storage technologies, offer a promising pathway toward deep carbon reductions. However, their nonlinear dynamics and operational uncertainty challenge traditional optimization methods. This paper presents a unified AI-enabled framework that combines forecasting models with a Reinforcement Learning (RL) controller for real-time, carbon-aware MES optimization. A comprehensive mathematical model is developed to capture multi-carrier interactions, storage dynamics, and time-varying carbon intensity. A Deep Deterministic Policy Gradient (DDPG) agent is trained using Long Short-Term Memory (LSTM) forecasts to minimize operational cost and emissions. Results from a smart-city case study show that the proposed approach reduces carbon emissions by 35.2%, lowers operational costs by 28.7%, and increases renewable utilization by 23.4% compared to rule-based control. The RL agent achieves performance within 4.8% of perfect-foresight optimization while maintaining robustness to forecasting errors. The framework demonstrates a scalable and adaptive pathway for intelligent, low-carbon energy management.
Index Terms—Artificial Intelligence, Low-Carbon Energy, Multi-Energy Systems, Smart Grid, Carbon Optimization.