Modeling and Dynamic Optimization in Oil Production
Abstract
The concept of Real-Time Production Optimization is a key for improving operational performance and productivity in the process industry. This includes utilization of computational technology, combining real-time measurements with mathematical models and optimization techniques, to offer a tool for decision support. In this thesis, dynamic optimization is applied to a subsea oil gathering network consisting of wells and flowlines, described by continuous-time nonlinear differential algebraic equations. The goal is to combine modeling and optimization tools to suggest operational conditions that optimize production over a short-term horizon. Three different approaches for optimal control are presented and assessed: direct collocation, single shooting and multiple shooting. These are implemented in the JModelica.org framework coupling state of the art numerical packages for modeling and optimization of dynamic systems. This includes IPOPT for solving nonlinear programs, CVODES providing integration of continuous-time ordinary differential equations with sensitivity information, CasADi offering efficient automatic differentiation and Modelica as a high-level modeling language.Experiments shows that direct collocation and single shooting methods are significantly faster than the multiple shooting implementation. A set of test scenarios for production optimization have been assessed, indicating an increase of $0.5-3.2 \%$ of production rates by tracking an unreachable reference, leading to a steady-state profile for production rates. When maximizing the objective function over 12 hours, the suggested solution increases the objective function value by $3.4-15 \%$.