Mixed-Integer Nonlinear Programming Heuristics Applied to a Shale Gas Production Optimization Problem

Abstract

Mixed-Integer nonlinear programs(MINLPs) are a general class of nonlinear optimization problems that have a wide array of real-world applications. These problems are in general notoriously difficult to solve, and it is therefore of great interest to develop heuristics that can aid the solution process. This thesis contains two major parts.In the first part, an objective feasibility pump, a heuristic for finding high quality feasible solutions of MINLPs is developed and implemented in the open source C++ project BONMIN. A computational study of this heuristic revealed that it is generally not more effective compared to other heuristics, but that it can be tailored to specific problems to yield improvements over other heuristics with regards to objective value. In the second part, extensions of a complex, dynamic shale gas production optimization problem are described and a simple heuristic for this problem is developed. A set of test problems is used to perform a benchmark study of the impact of using heuristics on this problem. The results of this study revealed that the new heuristics outperform other currently available heuristics and can find good feasible solutions in a fraction of the CPU time required by the default branch-and-bound solver in BONMIN.