東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

FindBook

Google Book

Amazon

博客來

Comparison of Reinforcement Learning Algorithms Applied to High-Fidelity Building Models.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Comparison of Reinforcement Learning Algorithms Applied to High-Fidelity Building Models./
作者:	Dey, Sourav.
面頁冊數:	1 online resource (133 pages)
附註:	Source: Masters Abstracts International, Volume: 83-04.
Contained By:	Masters Abstracts International83-04.
標題:	Architectural engineering. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28714086click for full text (PQDT)
ISBN:	9798460422067

Comparison of Reinforcement Learning Algorithms Applied to High-Fidelity Building Models.
Dey, Sourav.

Comparison of Reinforcement Learning Algorithms Applied to High-Fidelity Building Models. - 1 online resource (133 pages)

Source: Masters Abstracts International, Volume: 83-04.

Thesis (M.S.)--University of Colorado at Boulder, 2021.

Includes bibliographical references

Reinforcement learning has been shown to be a promising approach to sequential decision making after its recent success in the autonomous vehicles, robotics, marketing, and gaming industries. Reinforcement learning has gained traction due to the advancement in deep learning. It has enabled RL to scale to decision making problems in high dimensional state and action spaces. With its recent success, RL has also attracted the attention in the building automation and controls field. Building automation and controls is responsible for maintaining a comfortable, safe, and healthy indoor environment in an energy efficient way. Building controls have become more complicated and need to balance the trade-off between multiple goals of occupant comfort, energy efficiency, and the provision of grid flexibility. This set of competing operational objectives is difficult to balance with conventional rule based feedback controls. The application of reinforcement learning in advanced building controls is both emerging and promising due to the recent availability of rich building data, higher computational resources, while avoiding the time to develop and calibrate a controls model for every individual building and energy system. This thesis compares the performance of an online policy-gradient based method with an offline value-based reinforcement learning method applied to a high fidelity commercial building control problem. The two algorithms chosen for this purpose are Deep Q Network (DQN) and Proximal Policy Optimization (PPO), which are among the most popular reinforcement learning algorithms. DQN has been found to be successful in the building controls arena for quite some time now. Conversely, the PPO algorithm is relatively new and thus fewer studies exist related to the application of PPO to building controls. An OpenAI Gym interface is developed here for the BOPTEST framework, which is an open-source advanced building controls test bed developed by the International Building Performance Simulation Association. The building model used here is a high-fidelity Spawn of EnergyPlus model, which is a combination of an EnergyPlus envelope model with a detailed Modelica model for HVAC systems and components. The aim of this research is to evaluate the difference in performance of the two algorithms and also to provide recommendations for the design of the observed states, the reward shaping, the duration and selection of the training (sample efficiency), the control time step, and several hyperparameter values providing the desired output in the context of building controls.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2023

Mode of access: World Wide Web

ISBN: 9798460422067Subjects--Topical Terms:

3174102
Architectural engineering.
Subjects--Index Terms:

Building controlsIndex Terms--Genre/Form:

542853
Electronic books.

Comparison of Reinforcement Learning Algorithms Applied to High-Fidelity Building Models.
LDR:03922nmm a2200373K 4500 001 2356087
005 20230612071814.5
006 m o d
007 cr mn ---uuuuu
008 241011s2021 xx obm 000 0 eng d
020 $a 9798460422067
035 $a (MiAaPQ)AAI28714086
035 $a AAI28714086
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Dey, Sourav. $3 3696557
245 1 0 $a Comparison of Reinforcement Learning Algorithms Applied to High-Fidelity Building Models.
264 0 $c 2021
300 $a 1 online resource (133 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Masters Abstracts International, Volume: 83-04.
500 $a Advisor: Henze, Gregor.
502 $a Thesis (M.S.)--University of Colorado at Boulder, 2021.
504 $a Includes bibliographical references
520 $a Reinforcement learning has been shown to be a promising approach to sequential decision making after its recent success in the autonomous vehicles, robotics, marketing, and gaming industries. Reinforcement learning has gained traction due to the advancement in deep learning. It has enabled RL to scale to decision making problems in high dimensional state and action spaces. With its recent success, RL has also attracted the attention in the building automation and controls field. Building automation and controls is responsible for maintaining a comfortable, safe, and healthy indoor environment in an energy efficient way. Building controls have become more complicated and need to balance the trade-off between multiple goals of occupant comfort, energy efficiency, and the provision of grid flexibility. This set of competing operational objectives is difficult to balance with conventional rule based feedback controls. The application of reinforcement learning in advanced building controls is both emerging and promising due to the recent availability of rich building data, higher computational resources, while avoiding the time to develop and calibrate a controls model for every individual building and energy system. This thesis compares the performance of an online policy-gradient based method with an offline value-based reinforcement learning method applied to a high fidelity commercial building control problem. The two algorithms chosen for this purpose are Deep Q Network (DQN) and Proximal Policy Optimization (PPO), which are among the most popular reinforcement learning algorithms. DQN has been found to be successful in the building controls arena for quite some time now. Conversely, the PPO algorithm is relatively new and thus fewer studies exist related to the application of PPO to building controls. An OpenAI Gym interface is developed here for the BOPTEST framework, which is an open-source advanced building controls test bed developed by the International Building Performance Simulation Association. The building model used here is a high-fidelity Spawn of EnergyPlus model, which is a combination of an EnergyPlus envelope model with a detailed Modelica model for HVAC systems and components. The aim of this research is to evaluate the difference in performance of the two algorithms and also to provide recommendations for the design of the observed states, the reward shaping, the duration and selection of the training (sample efficiency), the control time step, and several hyperparameter values providing the desired output in the context of building controls.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2023
538 $a Mode of access: World Wide Web
650 4 $a Architectural engineering. $3 3174102
650 4 $a Computer science. $3 523869
650 4 $a Artificial intelligence. $3 516317
653 $a Building controls
653 $a High fidelity models
653 $a Reinforcement learning
655 7 $a Electronic books. $2 lcsh $3 542853
690 $a 0462
690 $a 0984
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 783688
710 2 $a University of Colorado at Boulder. $b Civil, Environmental, and Architectural Engineering. $3 3350108
773 0 $t Masters Abstracts International $g 83-04.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28714086 $z click for full text (PQDT)