A Computational Performance Comparison of MILP vs. MINLP Formulations for Oil Production Optimisation

Emmanuel Epelle, Dimitrios Gerogiorgis

Research output: Contribution to journalArticlepeer-review

Abstract

Model-based oil production systems optimisation under pressure and facility routing constraints is a testing challenge, especially in presence of complex downhole wellbore phenomena (water coning, slugging, phase separation). Nonlinearities and nonconvexities from underlying physics and binary
decisions exacerbate model complexity, yielding Mixed Integer Nonlinear Programs (MINLP). To guarantee solvability of optimisation formulations and reduce MINLP complexity, piecewise linearisation techniques based on Special Ordered Sets of type 2 (SOS2) constraints are developed towards approximating nonlinear functions and transforming models to Mixed Integer Linear Programs
(MILP). Nevertheless, computational analyses of MILP vs. MINLP formulations for oil production optimisation are scarce. This study explores the benefits of an MILP reformulation applied to three case studies of varying complexity. We compare MILP model results to original MINLP formulation solutions with multiple solvers, evaluating the impact of the number of linearisation breakpoints used on solution time, accuracy, robustness, model development effort and ease of automation.
Original languageEnglish
Article number106903
JournalComputers and Chemical Engineering
Volume140
Early online date22 May 2020
DOIs
Publication statusPublished - 2 Sep 2020

Fingerprint Dive into the research topics of 'A Computational Performance Comparison of MILP vs. MINLP Formulations for Oil Production Optimisation'. Together they form a unique fingerprint.

Cite this