|
|
|
|
|
A reservoir of knowledge Reservoir models are quite complicated. They account for different fluids like water, oil, and gas, different rock properties, the features of pumps and production wells, and sometimes complicated chemical reactions. To make modeling these complex systems computationally tractable, mathematicians subdivide the reservoir into a mesh of blocks. They then associate wells, pumps, and other equipment with individual blocks and solve an approximate model of the systems' fluid dynamics. The team uses IPARS, a multi-model, multi-phase reservoir simulator developed at the Center for Subsurface Modeling under the direction of Mary Wheeler. The output of IPARS is translated into production rates and ever-important revenue levels. Equipment is then moved around the mesh in order to compare different configurations and to find the best one. The possibilities aren't endless, but their sheer number sometimes makes researchers pine for the days of hunch-based prospecting. "You can have billions of possible configurations that need to be examined, so you can't just do an exhaustive search of the parameter space [the collection of all possible configurations within a given grid]," according to Tahsin Kurc, an assistant professor at Ohio State and part of the Multiscale Computing Lab that is led by Joel Saltz. A single IPARS run usually takes hours. If it's really difficult, hours can bleed into days. "Complexity usually translates into precision," Kurc says. "We want to move intelligently through the search space." Intelligent movement relies on a dynamic, data-driven optimization system. Large volumes of data obtained from earlier simulations and dynamically updated by new simulations or experimental measurements are stored, queried, and analyzed to find promising initial configurations. These configurations are then refined with on-the-fly monitoring and steering of the simulation and optimization processes. A set of simulations provides a rough sampling of the search space. Middleware tools from Saltz's team, called STORM and DataCutter, manage the very large amounts of data produced by these simulations. These tools are also used to identify good starting points for more comprehensive searches. Dynamic steering and collaboration tools--AutoMate and DISCOVER from Associate Professor Manish Parashar's lab at Rutgers--allow on-the-fly searches within these subsections. Sophisticated optimization algorithms coupled with IPARS models guide these searches by comparing configurations in the subsections.
|
 Visualization of oil reservoir simulation. Pumps, represented by asterisks, push oil toward wells, represented by small white circles, that draw oil from the ground. Blue areas indicate areas of high water concentration. Brown areas indicate areas of high oil concentration. |
||||||
