서브메뉴
검색
Statistical Machine Learning-Based Predictive Control of Nonlinear Processes
Statistical Machine Learning-Based Predictive Control of Nonlinear Processes
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152800
- ISBN
- 9798383698105
- DDC
- 621.3
- 서명/저자
- Statistical Machine Learning-Based Predictive Control of Nonlinear Processes
- 발행사항
- [Sl] : University of California, Los Angeles, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 249 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-02, Section: B.
- 주기사항
- Advisor: Christofides, Panagiotis D.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Los Angeles, 2024.
- 초록/해제
- 요약Data are an essential factor in the fourth industrial revolution, demanding engineers and scientists to leverage and analyze their potential for significantly improving the efficiency of industrial processes and their control systems. In classical industrial process control systems, the models are constructed using linear data-driven approaches, where parameters are adjusted based on experimental or simulated data. In certain critical control loops focused on optimizing profits, first-principles models are used to describe the fundamental physico-chemical phenomena, incorporating a small set of parameters derived from industrial or simulation data. However, despite the effectiveness of these classical modeling methods in many studies, there persists a significant challenge when modeling large-scale, complex non-linear systems within the field of process engineering. Traditional approaches often fall short of accurately representing the complexities and nonlinear dynamics inherent in large-scale industrial processes. Therefore, there are continuous efforts to conduct extensive studies on effective tools for model development and evaluation techniques. This is crucial because process models play a central role in advanced control strategies, particularly, model-based control systems such as model predictive control (MPC) and economic MPC (EMPC) frameworks. Therefore, accurate construction and evaluation of these models will contribute to achieving the desired performance and ensuring operational efficiency, ultimately leading to robust and reliable control systems.Machine learning techniques have proven to be an effective modeling tool in many engineering applications. More specifically, machine learning models have been used to model large-scale, complex nonlinear systems. These models are then integrated into MPC to achieve closed-loop stability. Among the many types of machine learning techniques, recurrent neural networks (RNNs) are widely used to model nonlinear processes involving time series data. This is due to their special structure, which allows useful previous information to be retained. In addition to complexities arising from nonlinearities and the large-scale nature of practical industrial processes, and challenges in modeling these systems, time delays pose significant challenges in nonlinear control systems. These delays can arise due to various sources such as transportation lags, sensor and actuator response times. Such delays can lead to instability, oscillations, and overall degradation in the performance of the control system. Hence, addressing these delays is crucial for maintaining the system's stability and optimizing its performance. Besides time-delay systems, there are also systems that experience different time-scale multiplicity, known as two-time scale systems. These types of systems require specific techniques to handle and design efficient model-based controllers to achieve closed-loop stability. Additionally, this dissertation includes an assessment of generalization error bounds for different types of machine learning models to evaluate their performance and reliability in various scenarios.In response to the factors highlighted, this dissertation presents the integration of machine learning techniques with model predictive control to stabilize the dynamics of nonlinear chemical processes. The dissertation begins with a comprehensive overview of its motivation, background, and structure. Then, it discusses the use of machine learning models within a model predictive control framework to stabilize a nonlinear system with time-delays. Additionally, the design of a machine learning-based predictor to compensate the effect of inputs delays is discussed. The closed-loop stability of the system achieved with Lyapunov-based model predictive controllers is investigated through theoretical analysis. Subsequently, a theoretical framework for deriving generalization error bounds for RNNs, partially connected recurrent RNNs, and long short-term memory (LSTM) RNNs are introduced. Next, we study generalization error bounds for models capturing the dynamics of two-time-scale systems and present simulation studies to address the modeling criteria of these systems under MPC frameworks, along with the necessary assumptions to achieve closed-loop stability. Throughout the dissertation, control methods are validated through their application in numerical simulations of nonlinear chemical processes, highlighting their effectiveness, performance and reliability.
- 일반주제명
- Electrical engineering
- 일반주제명
- Computer engineering
- 일반주제명
- Computer science
- 일반주제명
- Information technology
- 키워드
- Machine learning
- 기타저자
- University of California, Los Angeles Electrical and Computer Engineering 0333
- 기본자료저록
- Dissertations Abstracts International. 86-02B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017163839
■00520250211152800
■006m o d
■007cr#unu||||||||
■020 ▼a9798383698105
■035 ▼a(MiAaPQ)AAI31556305
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621.3
■1001 ▼aAlnajdi, Aisha M. A. S.
■24510▼aStatistical Machine Learning-Based Predictive Control of Nonlinear Processes
■260 ▼a[Sl]▼bUniversity of California, Los Angeles▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a249 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-02, Section: B.
■500 ▼aAdvisor: Christofides, Panagiotis D.
■5021 ▼aThesis (Ph.D.)--University of California, Los Angeles, 2024.
■520 ▼aData are an essential factor in the fourth industrial revolution, demanding engineers and scientists to leverage and analyze their potential for significantly improving the efficiency of industrial processes and their control systems. In classical industrial process control systems, the models are constructed using linear data-driven approaches, where parameters are adjusted based on experimental or simulated data. In certain critical control loops focused on optimizing profits, first-principles models are used to describe the fundamental physico-chemical phenomena, incorporating a small set of parameters derived from industrial or simulation data. However, despite the effectiveness of these classical modeling methods in many studies, there persists a significant challenge when modeling large-scale, complex non-linear systems within the field of process engineering. Traditional approaches often fall short of accurately representing the complexities and nonlinear dynamics inherent in large-scale industrial processes. Therefore, there are continuous efforts to conduct extensive studies on effective tools for model development and evaluation techniques. This is crucial because process models play a central role in advanced control strategies, particularly, model-based control systems such as model predictive control (MPC) and economic MPC (EMPC) frameworks. Therefore, accurate construction and evaluation of these models will contribute to achieving the desired performance and ensuring operational efficiency, ultimately leading to robust and reliable control systems.Machine learning techniques have proven to be an effective modeling tool in many engineering applications. More specifically, machine learning models have been used to model large-scale, complex nonlinear systems. These models are then integrated into MPC to achieve closed-loop stability. Among the many types of machine learning techniques, recurrent neural networks (RNNs) are widely used to model nonlinear processes involving time series data. This is due to their special structure, which allows useful previous information to be retained. In addition to complexities arising from nonlinearities and the large-scale nature of practical industrial processes, and challenges in modeling these systems, time delays pose significant challenges in nonlinear control systems. These delays can arise due to various sources such as transportation lags, sensor and actuator response times. Such delays can lead to instability, oscillations, and overall degradation in the performance of the control system. Hence, addressing these delays is crucial for maintaining the system's stability and optimizing its performance. Besides time-delay systems, there are also systems that experience different time-scale multiplicity, known as two-time scale systems. These types of systems require specific techniques to handle and design efficient model-based controllers to achieve closed-loop stability. Additionally, this dissertation includes an assessment of generalization error bounds for different types of machine learning models to evaluate their performance and reliability in various scenarios.In response to the factors highlighted, this dissertation presents the integration of machine learning techniques with model predictive control to stabilize the dynamics of nonlinear chemical processes. The dissertation begins with a comprehensive overview of its motivation, background, and structure. Then, it discusses the use of machine learning models within a model predictive control framework to stabilize a nonlinear system with time-delays. Additionally, the design of a machine learning-based predictor to compensate the effect of inputs delays is discussed. The closed-loop stability of the system achieved with Lyapunov-based model predictive controllers is investigated through theoretical analysis. Subsequently, a theoretical framework for deriving generalization error bounds for RNNs, partially connected recurrent RNNs, and long short-term memory (LSTM) RNNs are introduced. Next, we study generalization error bounds for models capturing the dynamics of two-time-scale systems and present simulation studies to address the modeling criteria of these systems under MPC frameworks, along with the necessary assumptions to achieve closed-loop stability. Throughout the dissertation, control methods are validated through their application in numerical simulations of nonlinear chemical processes, highlighting their effectiveness, performance and reliability.
■590 ▼aSchool code: 0031.
■650 4▼aElectrical engineering
■650 4▼aComputer engineering
■650 4▼aComputer science
■650 4▼aInformation technology
■653 ▼aMachine learning
■653 ▼aModel predictive control
■653 ▼aNonlinear chemical processes
■653 ▼aLong short-term memory
■690 ▼a0544
■690 ▼a0489
■690 ▼a0984
■690 ▼a0464
■71020▼aUniversity of California, Los Angeles▼bElectrical and Computer Engineering 0333.
■7730 ▼tDissertations Abstracts International▼g86-02B.
■790 ▼a0031
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17163839▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


