A multi-agent reinforcement learning with markov decision process for integrated energy systems in low-carbon economic

Integrated Energy Systems (IES) are vital for achieving a low-carbon economy, but their dynamic, uncertain, and distributed nature challenges traditional centralized scheduling. This study proposes MARL-DRSPER-GACC, a novel framework integrating Multi-Agent Reinforcement Learning (MARL) with Markov Decision Processes (MDP) to model IES as a Multi-Agent System (MAS). Using the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, the framework optimizes scheduling through two key innovations: Dynamic Reward Shaping with Predictive Entropy Regularization (DRSPER) for enhanced exploration and robustness under uncertainty, and Graph-Adaptive Consensus Communication (GACC) for efficient agent coordination with 50 % reduced communication overhead. An adaptive carbon trading mechanism balances operational costs and emissions. Simulations demonstrate 7.4–32.4 % reductions in Total Operating Cost (TOC), 5.6–29.2 % reductions in Total Carbon Emissions (TCE), and 2.2–12.2 % improvements in System Efficiency (SE) compared to baselines, with 95 % confidence intervals ensuring reliability. These results validate the framework’s effectiveness for scalable, real-time, low-carbon IES scheduling, supporting sustainable energy management.