Policy Regularization in Model-Free Building Control via Comprehensive Approaches from Offline to Online Reinforcement Learning
Published in Ph.D. Dissertation, University of California San Diego, 2024
Recommended citation: Liu, Hsin-Yu. Policy Regularization in Model-Free Building Control via Comprehensive Approaches from Offline to Online Reinforcement Learning. Ph.D. Dissertation, University of California San Diego, Jun. 2024. https://escholarship.org/content/qt0b23889v/qt0b23889v_noSplash_60b699fcef9c1cfc9988c4c5c4249a14.pdf
This dissertation develops a novel policy regularization framework for reinforcement learning in HVAC control systems, focusing on safe and efficient operation in real-world settings. Contributions include methods for offline-to-online policy regularization, open-source building batch RL datasets for benchmarking, and empirical results demonstrating energy and performance improvements in building control tasks.
Key contributions:
- A comprehensive policy regularization framework applicable to offline, online, and offline-to-online RL.
- Release of the B2RL open-source building batch RL dataset to facilitate benchmarking.
- Empirical evaluation showing improved stability and sample-efficiency in HVAC control applications. ```markdown — title: “Policy Regularization in Model-Free Building Control via Comprehensive Approaches from Offline to Online Reinforcement Learning” collection: publications permalink: /publication/policy-regularization-dissertation date: 2024-06-01 venue: “Ph.D. Dissertation, University of California San Diego” citation: “Liu, Hsin-Yu. Policy Regularization in Model-Free Building Control via Comprehensive Approaches from Offline to Online Reinforcement Learning. Ph.D. Dissertation, University of California San Diego, Jun. 2024.” paperurl: ‘’ —
This dissertation develops a novel policy regularization framework for reinforcement learning in HVAC control systems, focusing on safe and efficient operation in real-world settings. Contributions include methods for offline-to-online policy regularization, open-source building batch RL datasets for benchmarking, and empirical results demonstrating energy and performance improvements in building control tasks.
Key contributions:
- A comprehensive policy regularization framework applicable to offline, online, and offline-to-online RL.
- Release of the B2RL open-source building batch RL dataset to facilitate benchmarking.
- Empirical evaluation showing improved stability and sample-efficiency in HVAC control applications.
```