Seminar: Graduate Seminar
Multi-Agent Reinforcement Learning for Modeling, Simulating, and Optimizing Energy Markets
Date:
September,28,2025
Start Time:
10:00 - 11:00
Location:
506, Zisapel Building
Add to:
Lecturer:
Matan Levy
Research Areas:
| This work studies how to model and optimize modern, hybrid electricity markets using multi-agent reinforcement learning (MARL). We formalize the interaction between a central System Operator (SO) and price-responsive grid-edge agents (GEAgents) that can produce, consume, and store energy, and whose strategic behavior affects both demand and system stability. The framework treats dispatch planning and tariff setting as SO actions and learns GEAgentsโ responses under uncertainty. To evaluate policies, we built Energy-Net, a modular simulator that cleanly separates physical grid dynamics from agent logic and is compatible with off-the-shelf RL/MARL algorithms. Using a day-ahead market case study, we compare linear vs. quadratic pricing and joint learning of SO and GEAgents. Results show that coordinated MARL policies can reduce reliance on expensive reserves and better align incentives, while highlighting brittleness to forecasting error and the need for risk-aware extensions. The contribution is a general MARL formulation for hybrid markets, a practical simulator, and empirical evidence that adaptive pricing and dispatch can improve efficiency and stability in decentralized grids.
student Under the supervision of Dr. Sara Keren. |