서브메뉴
검색
Advancing Distribution Automation Through Model-Based and Machine Learning Approaches
Advancing Distribution Automation Through Model-Based and Machine Learning Approaches
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105603
- ISBN
- 9798265400536
- DDC
- 621.319
- 저자명
- Chen, Zhengrong.
- 서명/저자
- Advancing Distribution Automation Through Model-Based and Machine Learning Approaches
- 발행사항
- [Sl] : Georgia Institute of Technology, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 164 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: A.
- 주기사항
- Advisor: Meliopoulos, A. P.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2024.
- 초록/해제
- 요약With the rise in distributed energy resources (DERs) and electric vehicles (EVs), electric power systems, especially distribution systems, are transforming into active systems with renewable and low-carbon energy resources. Shifting from passive to active is one of the most significant characteristics of distribution systems, allowing bidirectional power flow from DERs. Such change brings challenges to distribution systems, including growing complexity, heightened uncertainty, frequent voltage violations, dynamic load demand, and cybersecurity issues. With the development of advanced metering infrastructures (AMI), leveraging extensive sampled and historical data enables real-time monitoring and control to enhance the resilience of Active Distribution Networks (ADNs) under uncertainties and cyberattacks. This dissertation aims to advance distribution automation (DA) in terms of protection, control, and optimization to ensure secure, reliable, and resilient distribution system operation. Specifically, this dissertation explores advanced model-based and data-driven methodologies in this framework, including state estimation, fault diagnosis, voltage control, and load management.First, we propose a Dynamic State Estimation-Based Protection (EBP) scheme for addressing protection challenges posed by Inverter-Based Resources (IBRs). Unlike the legacy state estimation method, Dynamic State Estimation (DSE) uses two consecutive sample values to compute the estimation instead of steady-state current and voltage phasors. The object-oriented DSE algorithm requires high-fidelity device models, measurement models, and a DSE process. The main idea of EBP is to detect faults by checking the consistency between the measurements and the mathematical system model. The fault is identified by a low confidence level, suggesting an inconsistency between the measurements and the model. Simulation results demonstrate the benefits of the proposed EBP method, including simple settings, high accuracy, and fast response time.Once the fault is detected, a real-time fault diagnosis framework is proposed for fault classification and fault locating. We utilize a supervised Deep Learning (DL) model for real-time fault classification. Compared with other Machine Learning (ML) methods, this model has lower computational complexity and is more efficient for processing long sequences of real-time data. This advantage makes it more efficient for processing long sequences of real-time data. Moreover, it can automatically learn relevant features from sequential data, reducing the need for manual feature engineering. The main idea of the Dynamic State Estimation-Based Fault Locating (EBFL) algorithm is to treat the fault location as a state to obtain its best estimation. A non-linear weighted least square (WLS) estimator is applied for the state estimation process. Note that the DSE applications, including EBP and EBFL, use the dynamic device model of distribution systems, expressed in a standard syntax referred to as State Control and Parameter Algebraic Quadratic Companion Form (SCPAQCF).A model-free Vol-VAR Optimization (VVO) method via Robust Deep Reinforcement Learning (RDRL) algorithm is developed to enhance the voltage profile of distribution systems. Unlike a standard DRL, RDRL offers improved performance in unpredictable conditions and the ability to handle power injection uncertainties, where the uncertainties are considered adversarial attacks, making it highly effective in finding an optimal or suboptimal solution without solving complex min-max optimization problems. Moreover, we design the uncertainty sets acquired by conformal prediction as the inputs of actor-critic networks. This information provides robustness during training. Numerical results illustrate the effectiveness of the proposed RDRL methodology for the online application of the VVO problems with enhanced safety regions.
- 일반주제명
- Smart grid technology
- 일반주제명
- Demand side management
- 일반주제명
- Energy consumption
- 일반주제명
- Electric vehicles
- 일반주제명
- Decision making
- 일반주제명
- Neural networks
- 일반주제명
- Behavioral psychology
- 일반주제명
- Home economics
- 일반주제명
- Information technology
- 일반주제명
- Transportation
- 기본자료저록
- Dissertations Abstracts International. 87-05A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2024 us c eng d■001000017360667
■00520260202105603
■006m o d
■007cr#unu||||||||
■020 ▼a9798265400536
■035 ▼a(MiAaPQ)AAI32316027
■035 ▼a(MiAaPQ)GeorgiaTech76969
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621.319
■1001 ▼aChen, Zhengrong.
■24510▼aAdvancing Distribution Automation Through Model-Based and Machine Learning Approaches
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a164 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: A.
■500 ▼aAdvisor: Meliopoulos, A. P.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2024.
■520 ▼aWith the rise in distributed energy resources (DERs) and electric vehicles (EVs), electric power systems, especially distribution systems, are transforming into active systems with renewable and low-carbon energy resources. Shifting from passive to active is one of the most significant characteristics of distribution systems, allowing bidirectional power flow from DERs. Such change brings challenges to distribution systems, including growing complexity, heightened uncertainty, frequent voltage violations, dynamic load demand, and cybersecurity issues. With the development of advanced metering infrastructures (AMI), leveraging extensive sampled and historical data enables real-time monitoring and control to enhance the resilience of Active Distribution Networks (ADNs) under uncertainties and cyberattacks. This dissertation aims to advance distribution automation (DA) in terms of protection, control, and optimization to ensure secure, reliable, and resilient distribution system operation. Specifically, this dissertation explores advanced model-based and data-driven methodologies in this framework, including state estimation, fault diagnosis, voltage control, and load management.First, we propose a Dynamic State Estimation-Based Protection (EBP) scheme for addressing protection challenges posed by Inverter-Based Resources (IBRs). Unlike the legacy state estimation method, Dynamic State Estimation (DSE) uses two consecutive sample values to compute the estimation instead of steady-state current and voltage phasors. The object-oriented DSE algorithm requires high-fidelity device models, measurement models, and a DSE process. The main idea of EBP is to detect faults by checking the consistency between the measurements and the mathematical system model. The fault is identified by a low confidence level, suggesting an inconsistency between the measurements and the model. Simulation results demonstrate the benefits of the proposed EBP method, including simple settings, high accuracy, and fast response time.Once the fault is detected, a real-time fault diagnosis framework is proposed for fault classification and fault locating. We utilize a supervised Deep Learning (DL) model for real-time fault classification. Compared with other Machine Learning (ML) methods, this model has lower computational complexity and is more efficient for processing long sequences of real-time data. This advantage makes it more efficient for processing long sequences of real-time data. Moreover, it can automatically learn relevant features from sequential data, reducing the need for manual feature engineering. The main idea of the Dynamic State Estimation-Based Fault Locating (EBFL) algorithm is to treat the fault location as a state to obtain its best estimation. A non-linear weighted least square (WLS) estimator is applied for the state estimation process. Note that the DSE applications, including EBP and EBFL, use the dynamic device model of distribution systems, expressed in a standard syntax referred to as State Control and Parameter Algebraic Quadratic Companion Form (SCPAQCF).A model-free Vol-VAR Optimization (VVO) method via Robust Deep Reinforcement Learning (RDRL) algorithm is developed to enhance the voltage profile of distribution systems. Unlike a standard DRL, RDRL offers improved performance in unpredictable conditions and the ability to handle power injection uncertainties, where the uncertainties are considered adversarial attacks, making it highly effective in finding an optimal or suboptimal solution without solving complex min-max optimization problems. Moreover, we design the uncertainty sets acquired by conformal prediction as the inputs of actor-critic networks. This information provides robustness during training. Numerical results illustrate the effectiveness of the proposed RDRL methodology for the online application of the VVO problems with enhanced safety regions.
■590 ▼aSchool code: 0078.
■650 4▼aSmart grid technology
■650 4▼aDemand side management
■650 4▼aEnergy consumption
■650 4▼aElectric vehicles
■650 4▼aDecision making
■650 4▼aNeural networks
■650 4▼aBehavioral psychology
■650 4▼aHome economics
■650 4▼aInformation technology
■650 4▼aTransportation
■690 ▼a0800
■690 ▼a0384
■690 ▼a0386
■690 ▼a0489
■690 ▼a0454
■690 ▼a0709
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05A.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360667▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


