Abstract
ExxonMobil’s proprietary reservoir simulator, EMpower, includes a programmable decision support system for manipulating well rates and wellhead pressures during simulation runs. The so-called well management component allows users to write algorithms that monitor simulation progress and steer the depletion process according to simulation objectives and constraints at well, platform, reservoir, and field levels. A built-in nonlinear optimization package can be used to determine well allocation under a set of constraints, including pressure and rate limits in the surface facility network. The strength of the system is its flexibility, which is limited only by the amount of programming a user is able and willing to do. Drawbacks of the system are its complexity, reliance on user expertise, and the requirement to write programs.
The move towards massively parallel computing platforms, such as clusters with thousands of nodes, poses a major challenge to the well management environment as we know it. Its inherently sequential programming model, which is geared towards simulation users rather than software developers, cannot take advantage of the massive parallelization that is expected to dominate simulation technology in the next decade, and could be a roadblock for parallel scalability.
In the following, we describe our efforts to adapt the system to a massive parallel simulation environment. At the heart of the transformation lies an approach that parallelizes constraint management and work-overs at the well level, and uses numerical optimization to control material balance, production, and injection for well-groups. After laying out the approach and describing the mathematical derivation of the optimization problem, we provide examples for scalability and parallel efficiency.
