MARLEM: A Multi-Agent Reinforcement Learning Simulation Framework for Implicit Cooperation in Decentralized Local Energy Markets
An open-source AI simulation trains energy agents to cooperate without communication, boosting grid stability.
Researchers Nelson Salazar-Pena, Alejandra Tabares, and Andres Gonzalez-Mancera built MARLEM, an open-source multi-agent reinforcement learning (MARL) framework. It simulates decentralized local energy markets (LEMs) as a Gymnasium environment. The key innovation enhances agents' observations with system-level KPIs, enabling them to learn cooperative strategies that benefit the entire grid—like optimizing battery storage—without direct communication. This allows researchers to test how market designs impact efficiency and stability in future energy systems.
Why It Matters
Provides a crucial tool for designing resilient, AI-driven smart grids that can balance supply, demand, and storage autonomously.