AI-enabled Automated Algorithm Selection and Configuration for Mathematical Optimization Problems

University Of Minnesota-Twin Cities, Minneapolis MN

Investigators

Abstract

Mathematical optimization is the cornerstone of decision-making in chemical engineering. In problems such as the design and real-time operation of chemical process systems interacting with renewable energy resources, the design of resilient supply chain networks, and the sustainable operation of chemical production facilities, the complex behavior of the underlying processes and the presence of multiple temporal and spatial scales lead to large-scale and complex optimization formulations. Algorithms for solving these problems have been and continue to be developed, but (i) their implementation is challenging and computationally intensive since they involve numerous steps, and (ii) it is not clear a-priori which algorithm is better suited for a given problem. In this research program, state-of-the-art artificial intelligence (AI) and machine learning (ML) tools will be employed to select and implement the best optimization algorithm for a given problem. These methods will be automated and incorporated in open-source software to facilitate the solution of complex problems by industry practitioners and academic researchers alike. In this research program, graduate students will be trained in fundamental research cutting across chemical engineering, mathematical optimization, and data science. Outreach activities to high schools in Minneapolis, rural Minnesota, and the American Indian, Hmong, and Somali communities in Minnesota will highlight the increasing importance of data science in the chemical industry and will aim to motivate careers in chemical engineering. This research will leverage state-of-the-art methods in artificial intelligence (AI) and machine learning (ML) to enable the automated selection and configuration of state-of-the-art optimization algorithms for the solution of nonlinear and mixed integer nonlinear problems that arise in process systems engineering. The research will address the following tasks: (i) a graph neural network framework will be developed to represent generic nonlinear optimization problems in a form that captures detailed information on the variables and constraints; (ii) geometric deep learning methods will be employed for the selection of the best solution strategy and the tuning of optimization algorithms in an automated manner; (iii) explainable AI methods will be employed to decode the relationship between optimization problems and the computational performance of optimization solvers. The optimization framework to be developed and implemented in open-source software will facilitate process systems researchers’ ability to solve complex decision-making problems efficiently by selecting the most appropriate solution method and optimally tuned solution algorithm. In addition, this framework will make possible detection of possible performance bottlenecks in current modeling practices and/or optimization algorithms, and in turn guide problem reformulation or algorithm improvements. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.