Abstract: The direct sampling stochastic simulation method for reservoir modeling was discussed, the selection of geological pattern component was improved, and a method was presented which combined the structural characteristic information of the spatial relationship with the pattern component. CUDA (compute unified device architecture)-based parallel strategies were also proposed for obtaining an optimal solution within the pattern subspaces. Experimental results show that the proposition of the pattern component selection greatly improves the large-scale continuity of the sand channels in the two-facies sedimentary system. Further, the parallel computing method for solving the pattern subspace has small time complexity. The parallel computational efficiency on GPU shows a 10X to 100X improvement compared with the serial implementation with different computing parameters.

Key words: stochastic simulation, direct sampling, parallel computing, compute unified device architecture (CUDA)