Towards a Generic Algorithm for Identifying High-Quality Decompositions of Optimization Problems

Abstract Optimization is a ubiquitous tool in process systems engineering which is being used on models of ever-increasing size and complexity. Decomposition solution approaches can be powerful tools for solving difficult optimization problems but are dependent on finding a partition of variables and constraints amenable to the solution approach. In this paper, we propose an automated, generic method for decomposing optimization problems using community detection. This method generates subproblems which are tightly interacting amongst themselves but weakly interacting with other subproblems, thus minimizing the amount of coordination required in the solution approach. We demonstrate the ability of our method to find solutions faster than solving the original problem in many cases.