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

Abstract

Abstract Reservoir description for simulation studies requires good knowledge of the permeabilities. Unfortunately, reliable permeability is only available from laboratory tests on cores, which are usually taken from a small percentage of the wells. Frequently, this information is extrapolated to calculate permeabilities all over the field, but the lack of enough data points usually causes unreliable predictions. We propose a method to estimate formation permeabilities from standard well logs and core data. The analysis includes a first step consisting of the interpretation of the petrophysics and a characterization in lithofacies, electrofacies and hydraulic flow units. This step involves the use of modern mathematical tools to rationally classify each reservoir region into a given (discrete) hydraulic flow unit. As a second step the core permeability data is mapped with the well log data using neural networks and the restrictions found on the first step of the analysis. This approach allows the use of continuous hydraulic flow unit values and reduces the error arising from the discrete zonation technique. Also, it overcomes the error coming from the mapping between log data and hydraulic flow units. The method should be applicable to any kind of reservoir as long as sufficient core and log data are available. The method assumes that the Carman-Kozeny equation holds for the reservoir rocks, which is a fairly reasonable assumption, and that the well logs available contain intrinsic information on tortuosity, sand size distribution, cementing characteristics, etc., which ultimately determine the flow performance of the rock. This hypothesis is usually strong because the available logs are not able to fully read the physical phenomena that cover the complex dynamics of the flow on the reservoir rocks. The method was tested using available core and log data in a sandstone formation in Chihuido de la Salina, Neuquen Basin, Argentina. Some core data points were not used to train the neural network and therefore useful for validation and comparison. In spite of the cited drawbacks, the method has shown to outperform both the standard regression techniques and the hydraulic flow units approach.