← Back to Projects
Datacenter Cooling with Deep Reinforcement Learning
January 1, 2024 • [completed]
RL agents for energy-efficient HVAC control. 94.4% water savings. Expanding to world models on 1.2B observations.
Tags: research, reinforcement-learning, datacenter, cooling, ddpg, energy, decision-transformers
Trained DDPG agents for datacenter cooling across multiple climate zones. 94.4% water savings in Seattle (+-0.5% variance across seeds), strong generalization to Phoenix, Chicago, and other climates. Sinergym simulation environment.
Expanding into territory where no published work exists: Decision Transformers (offline RL) and Dreamer-style world models on the EnergyBench dataset (1.2 billion observations). Also developing carbon-aware cooling optimization and multi-agent coordination for datacenter clusters.
Status: Initial research complete, expansion in progress
Stack: Python, Sinergym, Stable-Baselines3, DDPG/SAC/TD3, PyTorch
Related: "The Digital Dust Bowl" on Medium connects datacenter cooling to the Aral Sea crisis.