Abstract
dc:description.abstractRigorous computer simulation models are used routinely in chemical and other engineering and science disciplines to explain complex underlying phenomena. Solving optimization formulations with embedded computer simulations is of high interest since it bridges the gap between multi-fidelity and multiscale knowledge and quantitative decision-making. Due to the complexity of the simulation models, direct employment of equation-based optimization solvers may be impractical, and data-driven optimization techniques are often needed. Simulation-based optimization problems are often solved with black-box approaches, where the simulation is treated as an input-output data-generator. However, due to the absence of derivative information, existing black-box approaches often provide no convergence guarantees and require a significant number of samples to identify a solution with consistency. To tackle these challenges, we propose a novel data-driven spatial branch-and-bound algorithm (DDSBB) for simulation-based optimization problems, aiming for consistent globally convergent solutions. This dissertation begins by describing the basic structure of the DDSBB algorithm adopted from the equation-based branch-and-bound algorithms, which is a well-established field of study. We identify the essential building blocks to enable the transition from an equation-based algorithm to a data-driven algorithm: data-driven branching rules, data-driven underestimators, and constraint-handling techniques. The majority of the dissertation presents the development of theoretical and algorithmic strategies, including a support vector regression-based branching rule, two types of data-driven surrogate-based underestimators, a quadratic data-driven convex underestimator, and extensions to handle constraints. The rest of this dissertation documents focuses on the open-source software implementation of the DDSBB framework. The premise of the software implementation is to provide a user-friendly platform for simulation-based optimization. More importantly, this thesis consists of a variety of computational studies on illustrative examples and benchmark problems to show the efficacy of the algorithmic components and software implementation.
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Chemical and Biomolecular Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2021
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhai, Jianyuan
- Advisor dc:contributor.advisor
-
- Boukouvala, Fani
- Committee members dc:contributor.committeemember
-
- Grover, Martha
- Realff, Mathew
- Styczynski, Mark
- Dey, Santanu
Subjects
dc:subject × 3Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- http://hdl.handle.net/1853/66487