Mitrai, Ilias2023-11-282023-11-282023-07https://hdl.handle.net/11299/258630University of Minnesota Ph.D. dissertation. July 2023. Major: Chemical Engineering. Advisor: Prodromos Daoutidis. 1 computer file (PDF); xiii, 256 pages.Decision-making problems arise in a wide range of applications in chemical engineering. The efficient solution of complex and large-scale optimization problems is key for addressing problems related to the decarbonization of the chemical industry and the design and operation of sustainable manufacturing systems. Decomposition-based solution algorithms can be used to improve the tractability of such problems by exploiting their underlying structure. However, their application is problem-specific and time-consuming, and a generic framework for their automatic implementation is currently lacking. This thesis is focused on automating the application of decomposition-based solution algorithms and accelerating their performance using tools from artificial intelligence and network science, as discussed in the two parts of this thesis. The main contributions are: 1. We develop a graph classification approach, which can be applied to generic problems and considers the detailed coupling among the variables and constraints to determine whether an optimization problem should be solved using a decomposition-based or a monolithic solution approach. 2. We use Stochastic Blockmodeling (and its variants) and Bayesian inference to learn the underlying structure of the problem based on an appropriate graph representation of an optimization problem. The learned structure is used as the basis for the application of decomposition-based solution algorithms. 3. We propose a machine learning approach to learn how to optimally initialize cutting plane-based decomposition-based methods for the solution of optimization problems that arise in the real-time operation of chemical processes. 4. We explore the relation between problem formulation and the computational performance of decomposition-based solution methods and propose a new formulation and solution approach for the integration of process operations with dynamic optimization. 5. We propose a machine learning-based approach to accelerate Generalized Benders Decomposition for the solution of mixed integer model predictive control problems that arise in the operation of chemical processes.enAlgorithm configurationAlgorithm selectionArtificial IntelligenceDecomposition-based optimizationMathematical OptimizationThesis or Dissertation