Parallel Simulated Annealing for Stochastic Reservoir Modeling

Simulatedannealing (SA) Wutiques have shown great potential to generate geologically realistic permeability fields by combining data from many sources, such as well logs, cores, and tracer tests. However, the application of SA in reservoir description and simulation is limited owing to its prohibitively large computational time requirement, even on modern MqXWornputem. This paper introduces an implementation of a parallel SA algoxithmfor stochastic reservoir modeling on a Hypercube praxtwor network (Mel iPSC 860). The corresponding sequential code, which incorporates univariate and bivariate statistics to generate the permeability field, is optimized to gain maximum advantage fkom the parallel application. In this particularparallel implementation,each processorruns the same source code asynchronously using the single-instructionmultipledata (SXMD)approach, with synchronization linked to an optirmdity tesL By porting the SA algorithm to a parallel cOmputerwe can generatepermeabilityfielm that representnonGaussian bivariate statistics and complex flow geometry at a finer scale than was previously feasible.