A Method To Estimate Permeability on Uncored Wells Based on Well Logs and Core Data

摘要

Reservoir description for simulation studies requires goodrnknowledge of the permeabilities. Unfortunately, reliablernpermeability is only available from laboratory tests on cores,rnwhich are usually taken from a small percentage of the wells.rnFrequently, this information is extrapolated to calculaternpermeabilities all over the field, but the lack of enough datarnpoints usually causes unreliable predictions. We propose arnmethod to estimate formation permeabilities from standardrnwell logs and core data. The analysis includes a first steprnconsisting of the interpretation of the petrophysics and arncharacterization in lithofacies, electrofacies and hydraulic flowrnunits. This step involves the use of modern mathematical toolsrnto rationally classify each reservoir region into a givenrn(discrete) hydraulic flow unit. As a second step the corernpermeability data is mapped with the well log data usingrnneural networks and the restrictions found on the first step ofrnthe analysis. This approach allows the use of continuousrnhydraulic flow unit values and reduces the error arising fromrnthe discrete zonation technique. Also, it overcomes the errorrncoming from the mapping between log data and hydraulicrnflow units. The method should be applicable to any kind ofrnreservoir as long as sufficient core and log data are available.rnThe method assumes that the Carman-Kozeny equation holdsrnfor the reservoir rocks, which is a fairly reasonablernassumption, and that the well logs available contain intrinsicrninformation on tortuosity, sand size distribution, cementingrncharacteristics, etc., which ultimately determine the flowrnperformance of the rock. This hypothesis is usually strongrnbecause the available logs are not able to fully read thernphysical phenomena that cover the complex dynamics of thernflow on the reservoir rocks. The method was tested usingrnavailable core and log data in a sandstone formation inrnChihuido de la Salina, Neuquen Basin, Argentina. Some corerndata points were not used to train the neural network andrntherefore useful for validation and comparison. In spite of therncited drawbacks, the method has shown to outperform both thernstandard regression techniques and the hydraulic flowrnunits approach.