New transfer learning model enhances building control performance

Researchers have developed a novel zero-shot transfer learning approach for model predictive control (MPC) in buildings. This method utilizes generalized models pretrained on excitation-based operational data, which purposefully probes the building's state-action space. When applied without fine-tuning to 32 simulated target buildings, these excitation-based models demonstrated superior control performance, outperforming an online linear model-based MPC by 6.4% and a PI controller by 36.9%. This approach significantly reduces the cost and complexity of deploying MPC in the building sector by eliminating the need for target-specific data. AI

IMPACT Reduces setup costs and simplifies deployment of building control systems by eliminating the need for target-specific data.

RANK_REASON Academic paper detailing a new method for model predictive control using transfer learning. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.LG →

• arXiv
• cs.LG
• eess.SY
• Hugging Face
• model predictive control
• online linear model-based MPC
• transfer learning

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →