서브메뉴
검색
Computationally Incorporating Human and Climate Uncertainties in Energy System Planning
Computationally Incorporating Human and Climate Uncertainties in Energy System Planning
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202103211
- ISBN
- 9798314851272
- DDC
- 620
- 서명/저자
- Computationally Incorporating Human and Climate Uncertainties in Energy System Planning
- 발행사항
- [Sl] : Carnegie Mellon University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 349 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-11, Section: B.
- 주기사항
- Advisor: Nock, Destenie.
- 학위논문주기
- Thesis (Ph.D.)--Carnegie Mellon University, 2025.
- 초록/해제
- 요약In long-term energy system planning, both individual stakeholder preferences and future weather scenarios are two major sources of uncertainty that affect long-term energy system planning. Both of these sources of uncertainty also have their own modeling paradigms and unique challenges, especially when interfacing with existing energy system modeling and policy applications. In this dissertation, I present a body of work that demonstrates how different machine learning methods can be used to model individual preferences and representative future weather scenarios for energy system planning. This work also demonstrates a systematic evaluation of these methodological advances in the context of common difficulties in discrete choice elicitation. Chapter 2 presents a theoretical evaluation of parametric, semi-parametric, and nonparametric machine learning models for capturing individual discrete choice heterogeneity in the context of common challenges faced during policy stakeholder preference elicitation. Model performance depends on the context of the discrete choice paradigm, but increasing the abundance of choice sets and individual choice determinism improves model performance across all contexts. In general, semi- and nonparametric models outperform the parametric models more commonly used in policy applications.Chapter 3 extends the theoretical foundation of the machine learning model by combining it with a statistical preference recovery model and stated preference to form a novel three-stage revealed preference method for revealing an energy system stakeholder's preference for equality in energy system planning, as well as the decision attributes motivating the preference. It was found that stakeholders who value energy system equality prioritize electricity access to the least populated counties and the availability of certain grid technologies (solar PV minigrid and transmission lines), which often contrasts with their stated preference of valuing tiered electricity access at the population level. These findings are valuable for incorporating stakeholder preferences in expanding equality-centric energy system planning.Chapter 4 evaluates the feasibility of using cluster-based, temporally representative period selection methods to incorporate high-resolution future climate simulations into existing capacity expansion modeling paradigms. Across over 1000 capacity expansion simulations, we found that a k-medoids clustering-based representative period selection method, where each selected period is weighted by the cluster size to which it belongs to, outperforms all other clustering-based and tree-based selection methods because it is able to capture both the local and global representativeness of the selected periods. However, the impact of each method varies when the expansion plan is disaggregated at the technology level. We demonstrate the importance of evaluating representative selection methods in the context of capacity expansion outcomes, as the downstream impact of model selection results in billions of dollars in system cost and technology investment differences.Finally, the dissertation concludes by reiterating that data science, statistics, and machine learning algorithms can be viable tools for incorporating individual preferences and future climate dynamics into an equitable, sustainable, and reliable future grid, but that these methods should also be rigorously evaluated in the context of energy system planning and policy, as domain-specific evaluation heuristics are often insufficient when bridging these modeling paradigms.
- 일반주제명
- Engineering
- 일반주제명
- Public policy
- 일반주제명
- Energy
- 키워드
- Machine learning
- 키워드
- Uncertainty
- 기타저자
- Carnegie Mellon University Engineering and Public Policy
- 기본자료저록
- Dissertations Abstracts International. 86-11B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017357349
■00520260202103211
■006m o d
■007cr#unu||||||||
■020 ▼a9798314851272
■035 ▼a(MiAaPQ)AAI32001451
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620
■1001 ▼aCao, Sheng Lun Christine.▼0(orcid)0009-0006-0289-031X
■24510▼aComputationally Incorporating Human and Climate Uncertainties in Energy System Planning
■260 ▼a[Sl]▼bCarnegie Mellon University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a349 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-11, Section: B.
■500 ▼aAdvisor: Nock, Destenie.
■5021 ▼aThesis (Ph.D.)--Carnegie Mellon University, 2025.
■520 ▼aIn long-term energy system planning, both individual stakeholder preferences and future weather scenarios are two major sources of uncertainty that affect long-term energy system planning. Both of these sources of uncertainty also have their own modeling paradigms and unique challenges, especially when interfacing with existing energy system modeling and policy applications. In this dissertation, I present a body of work that demonstrates how different machine learning methods can be used to model individual preferences and representative future weather scenarios for energy system planning. This work also demonstrates a systematic evaluation of these methodological advances in the context of common difficulties in discrete choice elicitation. Chapter 2 presents a theoretical evaluation of parametric, semi-parametric, and nonparametric machine learning models for capturing individual discrete choice heterogeneity in the context of common challenges faced during policy stakeholder preference elicitation. Model performance depends on the context of the discrete choice paradigm, but increasing the abundance of choice sets and individual choice determinism improves model performance across all contexts. In general, semi- and nonparametric models outperform the parametric models more commonly used in policy applications.Chapter 3 extends the theoretical foundation of the machine learning model by combining it with a statistical preference recovery model and stated preference to form a novel three-stage revealed preference method for revealing an energy system stakeholder's preference for equality in energy system planning, as well as the decision attributes motivating the preference. It was found that stakeholders who value energy system equality prioritize electricity access to the least populated counties and the availability of certain grid technologies (solar PV minigrid and transmission lines), which often contrasts with their stated preference of valuing tiered electricity access at the population level. These findings are valuable for incorporating stakeholder preferences in expanding equality-centric energy system planning.Chapter 4 evaluates the feasibility of using cluster-based, temporally representative period selection methods to incorporate high-resolution future climate simulations into existing capacity expansion modeling paradigms. Across over 1000 capacity expansion simulations, we found that a k-medoids clustering-based representative period selection method, where each selected period is weighted by the cluster size to which it belongs to, outperforms all other clustering-based and tree-based selection methods because it is able to capture both the local and global representativeness of the selected periods. However, the impact of each method varies when the expansion plan is disaggregated at the technology level. We demonstrate the importance of evaluating representative selection methods in the context of capacity expansion outcomes, as the downstream impact of model selection results in billions of dollars in system cost and technology investment differences.Finally, the dissertation concludes by reiterating that data science, statistics, and machine learning algorithms can be viable tools for incorporating individual preferences and future climate dynamics into an equitable, sustainable, and reliable future grid, but that these methods should also be rigorously evaluated in the context of energy system planning and policy, as domain-specific evaluation heuristics are often insufficient when bridging these modeling paradigms.
■590 ▼aSchool code: 0041.
■650 4▼aEngineering
■650 4▼aPublic policy
■650 4▼aEnergy
■653 ▼aBehavioral heterogeneity
■653 ▼aEnergy system modeling
■653 ▼aMachine learning
■653 ▼aUncertainty
■690 ▼a0537
■690 ▼a0630
■690 ▼a0800
■690 ▼a0791
■71020▼aCarnegie Mellon University▼bEngineering and Public Policy.
■7730 ▼tDissertations Abstracts International▼g86-11B.
■790 ▼a0041
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17357349▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
