Preconditioned Newton methods for simulation of reservoirs with surface facilities

The primary focus of this research is to develop methods for increasing the computational efficiency of fully coupled implicit reservoir and surface facility problems, utilizing both sequential and parallel processing environments. This research also focuses on the object-oriented design of the application code with the goals of reusability and extendibility.; Explicit coupling is the most common method for simulation of reservoir flow with surface facilities and can result in significant material balance errors. A standard implicit formulation of the entire system will avoid material balance errors, however the number of Newton iterations generally increases. New methods for accelerating convergence of Newton's method for the fully coupled system are developed.; Two new techniques are introduced: (1) An adaptive explicit coupling method is introduced that combines the efficiency of explicit coupling with the accuracy of the standard implicit formulation. The method involves switching between fully implicit coupled reservoir and facilities and explicit coupling depending on the occurrence of phase transitions at the wells. The switching criteria developed is designed to employ the fully implicit facilities formulation during phase transitions. Outside the transition zone an explicit formulation is employed in order to avoid the expensive full Newton iteration. (2) A new preconditioning method is presented that accelerates convergence of the standard implicit facilities and adaptive explicit facilities formulations. The method utilizes domain decomposition techniques to accelerate Newton's method. Test results show that significant reductions in the number of Newton iterations can be obtained.; The most computationally expensive components of the preconditioning method are the facilities Jacobian calculations. Two parallel models are developed for distributing the Jacobian calculations across multiple processors. Each model is based on a different task assignment algorithm and good parallel efficiencies are observed for two to six processors.; The complexity of the application code developed for this research demonstrates the viability of object-oriented methods. Several examples are presented that demonstrate the reusability of the object data structures and computational methods. Additionally, the object and computational model design provide insights into the design of objects and methods required for the new preconditioning technique and parallel formulations.