Core Concepts
HVAC systems can achieve energy efficiency and occupant comfort by transitioning to interpretable decision tree policies and employing domain-specific verification criteria.
Abstract
This article discusses the transition from black-box models to decision tree policies for HVAC control optimization. It introduces verification criteria and highlights the benefits of the proposed approach through extensive experiments.
Abstract
- Model-based Reinforcement Learning (MBRL) potential for HVAC efficiency.
- Reliability concerns with existing methods.
- Redesigning HVAC controllers using decision trees for efficiency and reliability.
Introduction
- Efficient HVAC control crucial for energy consumption and occupant comfort.
- Model-free Reinforcement Learning (MFRL) explored for HVAC control.
- MBRL offers high data efficiency but faces reliability and interoperability challenges.
Proposed Approach
- Redesigning HVAC controllers using decision trees.
- Introducing verification criteria for HVAC control policies.
- Developing policy extraction procedure for decision tree policies.
Evaluation
- Assessing approach with high-fidelity simulator.
- Comparing performance with benchmarks.
- Demonstrating energy savings and improved comfort with the proposed method.
Stats
최신 방법과 비교하여 에너지 절약률은 68.4% 증가.
인간의 편안함 증가율은 14.8% 증가.
계산 오버헤드는 이전 최첨단 방법보다 1127배 빠름.
Quotes
"Our method saves 68.4% more energy and increases human comfort gain by 14.8%."
"Our method is significantly faster, reducing computation overhead by 1127 times."