High performance reservoir simulation on PC clusters

摘要

Reservoir simulation is an important and challengine problem for the oil and gas industry. A correctly calibrated reservoir simulator provides a cost-effective tool for reservoir evaluation that can be used to obtain essential reservoir information. A long-standing problem in reservoir simulation is history matching, which aims to search for a suitable set of values for the simulator's input parameters so that the simulator can correctly predict reservoir wells' fluid (oil, gas, etc.) outputs over a period of time. Due to the sheer size of the problem, completely satisfactory results of history matching have been hard to achieve and are usually obtained with ad hoc methods. Fortunately, the problem possesses inherent parallelism. This paper presents a novel technique of using fuzzy control and neural networks to improve history matching. When implemented on a cluster of PCs for "Web based parallel processing," our technique aims to solve history matching faster and at a lower cost, thus making history matching and reservoir modeling economically feasible for oil producers.