서브메뉴
검색
Human-In-The-Loop Energy-Efficient Building HVAC Control
Human-In-The-Loop Energy-Efficient Building HVAC Control
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105507
- ISBN
- 9798263326333
- DDC
- 519.5
- 서명/저자
- Human-In-The-Loop Energy-Efficient Building HVAC Control
- 발행사항
- [Sl] : Georgia Institute of Technology, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 188 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Zhang, Ying.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2024.
- 초록/해제
- 요약The Heating, Ventilation, and Air Conditioning (HVAC) systems are essential units in buildings that regulate indoor temperature, humidity, and air quality to ensure occupant comfort. They are also major energy-consuming appliances in building systems, accounting for almost 40% of the total energy consumption in the United States. On the one hand, reducing the energy consumption of HVAC systems is crucial for decreasing greenhouse gas emissions. On the other hand, it is important to maintain the comfort levels of occupants while implementing energy-saving HVAC controllers. The objective of this dissertation is to design human-in-the-loop energy-efficient controllers for building HVAC systems.The thermal dynamics model of a building system is highly nonlinear and difficult to describe with analytic equations. Therefore, traditional control methods that require a simple yet sufficiently accurate model cannot be used. To address this problem, we use Reinforcement Learning (RL) to develop HVAC controllers for environments with complex dynamics. The RL agent is trained in a data-driven manner using a dataset collected through interactions with the environment. Since the data collection process in a building system is typically slow, model-based RL is used to derive the HVAC controller, which is more sample-efficient than model-free RL algorithms. Additionally, our proposed method combines model-based RL and Model Predictive Control (MPC) to avoid compounding errors by imitating the results of MPC random shooting. We conduct simulation experiments in the EnergyPlus environment to demonstrate the effectiveness of the proposed algorithm, with comparisons to existing algorithms highlighting the advantages of our method in energy efficiency and indoor temperature regulation.Different occupants have varied thermal preferences, making it challenging for a shared HVAC controller to ensure personalized comfort for every individual. To address this, we employ a meta-learning algorithm to develop a personalized thermal comfort model with minimal feedback instances. Utilizing the learned meta-model, we create a method that leverages the backpropagation of Neural Networks (NNs) to determine the optimal environmental and personal conditions for each occupant. A notable benefit of this identification algorithm is that thermal comfort, as indicated by the mean thermal sensation vote, progressively improves during the data collection phase. We validate the effectiveness of the meta-learning algorithm using the American Society of Heating, Refrigerating and Airconditioning Engineers (ASHRAE) global thermal comfort database II, showing that it can enhance prediction accuracy after gathering just five thermal sensation votes from an occupant. Additionally, we demonstrate the identification algorithm's efficacy in finding the best personalized thermal environmental conditions through a thermal sensation generation model based on the Predicted Mean Vote (PMV) model.The final chapter of this dissertation focuses on further improving the RL-based HVAC controllers by leveraging offline datasets and the personalized thermal comfort model. The exploration actions when training an HVAC controller via online RL may lead to unsafe states, such as an unreasonably high indoor air temperature, which makes the occupants extremely uncomfortable. By designing a high-performance model-based offline RL algorithm, we aim to mitigate the requirement of on-policy data for training that the existing RL-based HVAC controllers typically have. We train an ensemble NN to predict the thermal states of the considered zone. The obtained ensemble networks can indicate the regions in the state and action spaces covered by the offline dataset. With the personalized thermal preference model updated via meta-testing, model-based RL is used to derive the optimal HVAC controller so that the obtained HVAC controller can adapt to different occupants' thermal preferences with minimal thermal feedback. Since the proposed algorithm only requires offline datasets and minimal online thermal feedback for training, it contributes to a safer training process and more practical deployment of the RL algorithm in HVAC systems. Numerical simulations in the EnergyPlus environment demonstrate that the proposed algorithm can guarantee personalized thermal preferences with a slight increase in power consumption of 1.91% compared to the model-based RL algorithm with on-policy data aggregation.In summary, the proposed methods presented in this dissertation achieve human-in-theloop energy-efficient control for HVAC systems that ensure personalized thermal comfort for occupants. The improved HVAC controllers are promising for applications in real building systems and may provide new practical technologies for smart buildings in the future.
- 일반주제명
- Mean square errors
- 일반주제명
- Humidity
- 일반주제명
- Greenhouse gases
- 일반주제명
- Air conditioning
- 일반주제명
- Neural networks
- 일반주제명
- Controllers
- 일반주제명
- HVAC
- 일반주제명
- Energy consumption
- 일반주제명
- Ventilation
- 일반주제명
- Markov analysis
- 일반주제명
- Climate change
- 일반주제명
- Sustainability
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2024 us c eng d■001000017360331
■00520260202105507
■006m o d
■007cr#unu||||||||
■020 ▼a9798263326333
■035 ▼a(MiAaPQ)AAI32308066
■035 ▼a(MiAaPQ)GeorgiaTech78602
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a519.5
■1001 ▼aChen, Liangliang.
■24510▼aHuman-In-The-Loop Energy-Efficient Building HVAC Control
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a188 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Zhang, Ying.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2024.
■520 ▼aThe Heating, Ventilation, and Air Conditioning (HVAC) systems are essential units in buildings that regulate indoor temperature, humidity, and air quality to ensure occupant comfort. They are also major energy-consuming appliances in building systems, accounting for almost 40% of the total energy consumption in the United States. On the one hand, reducing the energy consumption of HVAC systems is crucial for decreasing greenhouse gas emissions. On the other hand, it is important to maintain the comfort levels of occupants while implementing energy-saving HVAC controllers. The objective of this dissertation is to design human-in-the-loop energy-efficient controllers for building HVAC systems.The thermal dynamics model of a building system is highly nonlinear and difficult to describe with analytic equations. Therefore, traditional control methods that require a simple yet sufficiently accurate model cannot be used. To address this problem, we use Reinforcement Learning (RL) to develop HVAC controllers for environments with complex dynamics. The RL agent is trained in a data-driven manner using a dataset collected through interactions with the environment. Since the data collection process in a building system is typically slow, model-based RL is used to derive the HVAC controller, which is more sample-efficient than model-free RL algorithms. Additionally, our proposed method combines model-based RL and Model Predictive Control (MPC) to avoid compounding errors by imitating the results of MPC random shooting. We conduct simulation experiments in the EnergyPlus environment to demonstrate the effectiveness of the proposed algorithm, with comparisons to existing algorithms highlighting the advantages of our method in energy efficiency and indoor temperature regulation.Different occupants have varied thermal preferences, making it challenging for a shared HVAC controller to ensure personalized comfort for every individual. To address this, we employ a meta-learning algorithm to develop a personalized thermal comfort model with minimal feedback instances. Utilizing the learned meta-model, we create a method that leverages the backpropagation of Neural Networks (NNs) to determine the optimal environmental and personal conditions for each occupant. A notable benefit of this identification algorithm is that thermal comfort, as indicated by the mean thermal sensation vote, progressively improves during the data collection phase. We validate the effectiveness of the meta-learning algorithm using the American Society of Heating, Refrigerating and Airconditioning Engineers (ASHRAE) global thermal comfort database II, showing that it can enhance prediction accuracy after gathering just five thermal sensation votes from an occupant. Additionally, we demonstrate the identification algorithm's efficacy in finding the best personalized thermal environmental conditions through a thermal sensation generation model based on the Predicted Mean Vote (PMV) model.The final chapter of this dissertation focuses on further improving the RL-based HVAC controllers by leveraging offline datasets and the personalized thermal comfort model. The exploration actions when training an HVAC controller via online RL may lead to unsafe states, such as an unreasonably high indoor air temperature, which makes the occupants extremely uncomfortable. By designing a high-performance model-based offline RL algorithm, we aim to mitigate the requirement of on-policy data for training that the existing RL-based HVAC controllers typically have. We train an ensemble NN to predict the thermal states of the considered zone. The obtained ensemble networks can indicate the regions in the state and action spaces covered by the offline dataset. With the personalized thermal preference model updated via meta-testing, model-based RL is used to derive the optimal HVAC controller so that the obtained HVAC controller can adapt to different occupants' thermal preferences with minimal thermal feedback. Since the proposed algorithm only requires offline datasets and minimal online thermal feedback for training, it contributes to a safer training process and more practical deployment of the RL algorithm in HVAC systems. Numerical simulations in the EnergyPlus environment demonstrate that the proposed algorithm can guarantee personalized thermal preferences with a slight increase in power consumption of 1.91% compared to the model-based RL algorithm with on-policy data aggregation.In summary, the proposed methods presented in this dissertation achieve human-in-theloop energy-efficient control for HVAC systems that ensure personalized thermal comfort for occupants. The improved HVAC controllers are promising for applications in real building systems and may provide new practical technologies for smart buildings in the future.
■590 ▼aSchool code: 0078.
■650 4▼aMean square errors
■650 4▼aHumidity
■650 4▼aGreenhouse gases
■650 4▼aAir conditioning
■650 4▼aNeural networks
■650 4▼aControllers
■650 4▼aHVAC
■650 4▼aEnergy consumption
■650 4▼aVentilation
■650 4▼aMarkov analysis
■650 4▼aClimate change
■650 4▼aSustainability
■690 ▼a0729
■690 ▼a0800
■690 ▼a0404
■690 ▼a0796
■690 ▼a0640
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360331▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


