FairMarket-RL: LLM-driven fairness shaping in multi-agent energy trading for smart distribution grids

Peer-to-peer (P2P) trading is increasingly recognized as a key mechanism for decentralized market regulation, yet existing approaches often lack robust frameworks to ensure fairness. This paper presents FairMarket-RL, a novel hybrid framework that combines Large Language Models (LLMs) with Reinforcement Learning (RL) to enable fairness-aware trading agents. In a simulated P2P microgrid with multiple sellers and buyers, the LLM acts as a real-time fairness critic, evaluating each trading episode using two metrics: Fairness-To-Buyer (FTB) and Fairness-Between-Sellers (FBS). These fairness scores are integrated into agent rewards through scheduled λ-coefficients, forming an adaptive LLM-guided reward shaping loop that replaces brittle, rule-based fairness constraints. A linear λ ramping curriculum stabilizes training and gradually incorporates fairness with no loss in convergence. Agents are trained using Independent Proximal Policy Optimization (IPPO) and achieve equitable outcomes, fulfilling over 90% of buyer demand, maintaining fair seller margins, and consistently reaching FTB and FBS scores above 0.80. Notably, the framework demonstrates resilience against adversarial behaviors and noise, maintaining stable performance even with corrupted fairness inputs. The training process demonstrates that fairness feedback improves convergence, reduces buyer shortfalls, and narrows profit disparities between sellers. Because the language-based critic requires no hand-crafted fairness features, the framework is structured to extend to settings with larger numbers of prosumers, a direction we examine qualitatively and leave to future work for full validation. FairMarket-RL thus offers an equity-driven approach for autonomous trading in decentralized energy systems.