Reinforcement Learning for BESS
Date:
Presentation of my Master’s project on the application of Reinforcement Learning for the control of a Battery Energy Storage System (BESS) under uncertain electricity price and demand conditions.
The presentation introduced a custom Gymnasium-based simulation environment modeling battery dynamics such as State of Charge (SoC), State of Health (SoH), and battery degradation while incorporating uncertain electricity price and demand forecasts. The project focused on the two energy management scenarios Energy Arbitrage and Peak Shaving.
A rule-based baseline controller was compared with multiple RL agents (DQN, TD3, QR-DQN) implemented using Stable-Baselines3. The results showed that RL-based approaches achieved higher revenues in the arbitrage scenario, while DQN and QR-DQN demonstrated better peak-shaving performance and robustness against forecast uncertainty.
The presentation was given in German.
📥 Download the presentation