서브메뉴
검색
Stochastic Optimal Control: Threshold-Aware Policies and Impact of Random Disruptions
Stochastic Optimal Control: Threshold-Aware Policies and Impact of Random Disruptions
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152713
- ISBN
- 9798384053217
- DDC
- 519
- 저자명
- Wang, MingYi.
- 서명/저자
- Stochastic Optimal Control: Threshold-Aware Policies and Impact of Random Disruptions
- 발행사항
- [Sl] : Cornell University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 214 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-03, Section: B.
- 주기사항
- Advisor: Vladimirsky, Alexander.
- 학위논문주기
- Thesis (Ph.D.)--Cornell University, 2024.
- 초록/해제
- 요약Stochastic optimal control theory encompasses various types of stochasticity and notions of optimality. The standard risk-neutral approach minimizes or maximizes an expected total cost, but this approach often yields non-robust results. In this thesis, we introduce a particular type of robust control framework of indefinite-horizon processes, maximizing the probability of desired outcomes while keeping the cumulative cost within a threshold.For diffusive processes, our framework results in second-order parabolic Hamilton-Jacobi-Bellman (HJB) Partial Differential Equations (PDEs).We develop an efficient algorithm to solve these equations by leveraging the inherent causality of the framework. This allows us to recover the optimal "threshold (risk)-aware" feedback policies for all initial configurations and a range of threshold values simultaneously in a single sweep. We first apply this methodology to adaptive cancer therapy under stochastic cancer dynamics. In particular, we aim to maximize the probability of achieving treatment goals while keeping the total treatment cost within a specific cost threshold/budget.We then extend this threshold-aware approach to hybrid control problems, specifically through sailboat routing under stochastically evolving wind conditions. This application involves solving a pair of quasi-variational inequalities in a Hamilton-Jacobi framework. Monte Carlo simulations are used to generate cumulative distribution functions (CDFs), demonstrating the advantages of threshold-aware policies over risk-neutral ones.In the final section, we investigate bacterial competition influenced by environmental extreme events (dilutions). We propose an explanation for why toxin-sensitive bacteria, usually outcompeted by toxin-producers in vitro, can thrive under frequent dilutions. We consider both deterministic periodic dilutions and randomly timed dilutions modeled by a Poisson process. Through a series of optimized toxin-regulation behaviors for toxin-producers, we demonstrate that toxin-sensitive strains still have a reasonable chance of winning. The numerical approach involves solving Hamilton-Jacobi-type equations (including a specific type of non-local coupling emerging from the jump-discontinuities induced by the Poisson process) using semi-Lagrangian schemes.
- 일반주제명
- Applied mathematics
- 일반주제명
- Mathematics
- 일반주제명
- Computer science
- 일반주제명
- Ecology
- 키워드
- Hybrid systems
- 기타저자
- Cornell University Applied Mathematics
- 기본자료저록
- Dissertations Abstracts International. 86-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017163480
■00520250211152713
■006m o d
■007cr#unu||||||||
■020 ▼a9798384053217
■035 ▼a(MiAaPQ)AAI31488853
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a519
■1001 ▼aWang, MingYi.▼0(orcid)0009-0007-0488-9248
■24510▼aStochastic Optimal Control: Threshold-Aware Policies and Impact of Random Disruptions
■260 ▼a[Sl]▼bCornell University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a214 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-03, Section: B.
■500 ▼aAdvisor: Vladimirsky, Alexander.
■5021 ▼aThesis (Ph.D.)--Cornell University, 2024.
■520 ▼aStochastic optimal control theory encompasses various types of stochasticity and notions of optimality. The standard risk-neutral approach minimizes or maximizes an expected total cost, but this approach often yields non-robust results. In this thesis, we introduce a particular type of robust control framework of indefinite-horizon processes, maximizing the probability of desired outcomes while keeping the cumulative cost within a threshold.For diffusive processes, our framework results in second-order parabolic Hamilton-Jacobi-Bellman (HJB) Partial Differential Equations (PDEs).We develop an efficient algorithm to solve these equations by leveraging the inherent causality of the framework. This allows us to recover the optimal "threshold (risk)-aware" feedback policies for all initial configurations and a range of threshold values simultaneously in a single sweep. We first apply this methodology to adaptive cancer therapy under stochastic cancer dynamics. In particular, we aim to maximize the probability of achieving treatment goals while keeping the total treatment cost within a specific cost threshold/budget.We then extend this threshold-aware approach to hybrid control problems, specifically through sailboat routing under stochastically evolving wind conditions. This application involves solving a pair of quasi-variational inequalities in a Hamilton-Jacobi framework. Monte Carlo simulations are used to generate cumulative distribution functions (CDFs), demonstrating the advantages of threshold-aware policies over risk-neutral ones.In the final section, we investigate bacterial competition influenced by environmental extreme events (dilutions). We propose an explanation for why toxin-sensitive bacteria, usually outcompeted by toxin-producers in vitro, can thrive under frequent dilutions. We consider both deterministic periodic dilutions and randomly timed dilutions modeled by a Poisson process. Through a series of optimized toxin-regulation behaviors for toxin-producers, we demonstrate that toxin-sensitive strains still have a reasonable chance of winning. The numerical approach involves solving Hamilton-Jacobi-type equations (including a specific type of non-local coupling emerging from the jump-discontinuities induced by the Poisson process) using semi-Lagrangian schemes.
■590 ▼aSchool code: 0058.
■650 4▼aApplied mathematics
■650 4▼aMathematics
■650 4▼aComputer science
■650 4▼aEcology
■653 ▼aDynamic programming
■653 ▼aPartial Differential Equations
■653 ▼aHybrid systems
■653 ▼aPiecewise-deterministic process
■653 ▼aStochastic optimal control
■653 ▼aThreshold (risk)-awareness
■690 ▼a0364
■690 ▼a0405
■690 ▼a0984
■690 ▼a0329
■71020▼aCornell University▼bApplied Mathematics.
■7730 ▼tDissertations Abstracts International▼g86-03B.
■790 ▼a0058
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17163480▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


