Using Visualization and Parallel Computing for Interactive Reservoir Characterization

Abstract With the application of parallel computing to reservoir modeling, namely in geostatistics and in flow simulations, there is an increasing need to handle and visualize complex 3-D volumes of reservoir descriptions. In this paper, we describe an approach to inspect and analyze multiple geostatistical descriptions of a reservoir using modular visualization and immersive virtual environments. In this approach, constrained geostochastic simulations are performed in parallel on an IBM SP2 and displayed in real time either on a graphic workstation or in an immersive environment. The rendering and volume manipulation is performed using a modular visualization environment (IBM Data Explorer). Using coarse grained parallelism, the realizations are renderedin a distributed computing environment by defining execution groups of visualization tasks that can be created or modified interactively through a visual modular program. After the definition of the execution groups, each group can be assigned to a set of SP nodes or workstations. The geostatistical realizations are transferred to the execution groups of the visualization environment by using message passing (PVMIMPI). In addition, to eliminate large traffic of data through the network. the images can be rendered locally on the nodes and only the images are transferred to the graphics workstations. Preliminary results have shown that this approach can be usehl for a group of geologists, geophysicists and reservoir engineers to quickly inspect multiple realizations simultaneously and to discard realizations that are statistically probable but not geologically tenable or not commensurate with the production field data. Introduction In the management of reservoirs, the prediction of reservoirperformance by flow simulations is valuable for the economic evaluation of producing fields. However, to obtain accurate predictions, realistic reservoir descriptions are required. Recently, stochastic simulations have been widely used both to generate realistic reservoir images and to assess the uncertainty associated with available data. This uncertainty is related to he relatively small amount of available data in comparison to the whole volume of the reservoir. Detailed reservoir description grids may contain several million cells, and stochastic simulations of these fine resolution grids are typically executed in non-interactive fashion because of the large computational resources required. Amoco recently ran a case study on a newly developed parallel flow simulator .in whlch 50 simulations were performed with more than and a half million cells per description1. In their study, among several suggestions to improve the cycle time of the work flow in reservoir modeling, they pointed out that (1) the process of generation of geostatistical realizations could be more automated; (2) statistical measures of the realizations could be more easily obtained and (3) integrated software environments could be used to allow seamless integration and automation of the entire work process cycle. In this paper, we address these problems by prototyping a computational environment for interactive reservoir characterization using a combination of several technologies, including visual programming, high performance parallel computing and distributed visualization