Multi-agent Reinforcement Learning-based Joint Design of Low-Carbon P2P Market and Bidding Strategy in Microgrids

A research team led by Junhao Ren has developed a groundbreaking AI framework to optimize energy trading within microgrid communities. The system tackles the dual challenge of renewable energy uncertainty and real-time market instability by creating an intraday P2P trading platform. Unlike traditional centralized optimization methods, this approach uses a Multi-Agent Reinforcement Learning (MARL) model where each microgrid acts as an autonomous, self-interested agent. Their decision-making is formulated as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP), allowing them to independently pursue economic benefits through strategic bidding.

The key innovation is a novel market-clearing mechanism that provides macro-regulation. While microgrids compete for profit, this overarching mechanism incentivizes behaviors that maximize social welfare—specifically, prioritizing local renewable consumption to reduce carbon emissions. Simulation results demonstrate the framework successfully balances individual economic pursuits with community-wide environmental goals, leading to a significant decrease in reliance on high-carbon external electricity. This represents a major shift from restrictive coordination rules toward a more dynamic, AI-driven market design that could make clean energy microgrids more efficient and scalable in real-world applications.