Log-Derived Permeability in a Heterogeneous Carbonate Reservoir of Middle East, Abu Dhabi, Using Artificial Neural Network

Abstract

Abstract Estimation of permeability in carbonates has been a challenge for many years. Well logs, particularly high-resolution logs, are influenced by rock properties. Therefore, when there is limited core coverage and scarce high-resolution log data, permeability estimation using the standard suite of logs (resistivity, density, neutron, caliper, gamma ray) is crucial for populating and constraining a 3D geological permeability model. Two new traces, the deep and micro resistivity activity traces, are derived from the corresponding resistivity logs. The activity traces are not affected by fluid effects and, thus, preserve better the formation characteristics. Permeability estimation using an artificial neural network approach is made through a two-step process. In the first step, probabilities of log-derived rock types are estimated from a trained neural network using the micro and deep resistivity activity traces, and the standard suite of logs as input. In the second step, a separately trained neural network uses rock type probabilities from step 1, along with a suite of logs to predict permeability. Two examples are provided to illustrate the validity of the method in predicting permeability in a heterogeneous carbonate reservoir located in Abu Dhabi, UAE. This reservoir exhibits permeability ranging from half a milli-Darcy to more than 20 Darcies. The first example represents a blind test where the estimated permeability shows good agreement with core permeability data. The second example demonstrates the predictive capability of the method in a non-cored well that is located in the vicinity of cored wells. The estimation technique is robust and was found valuable to supplement core data in the construction of geo-cellular permeability models. Introduction Many challenges exist in characterizing, predicting, and quantifying carbonate reservoir quality. Carbonate rocks are susceptible to modification by post-depositional mechanisms. Processes such as compaction, lithification, dolomitization, and others result in large variations in the reservoir quality of carbonates. Carbonates are characterized by different porosity types with complex pore size distributions, which result in wide permeability variations for a given level of total porosity. In other words, there is no simple relationship between porosity and permeability. This makes it difficult to predict the rock fluid flow characteristics. Responses of logs vary according to rock properties. Evaluation of various aspects of carbonates requires a special suite of logs and interpretation techniques, in order to predict fluid flow properties. Permeability is the most important property that controls fluid flow in porous media. Direct measurement of permeability, as a log has not been accomplished. For carbonate reservoirs, where there is no simple relationship between porosity and permeability, the petrophysicist faces the difficult task of relating the measured log properties to core permeability. Once having accomplished this task, the petrophysicist can attempt to predict permeability for non-cored wells. Historically, permeability has been estimated using porosity-permeability transforms generated through regression of core porosity and permeability data. This technique is considered adequate for sandstone reservoirs where the pore geometry can be predicted based on their depositional environments. These relationships, however, fail in predicting permeability in complex carbonates where digenetic processes introduce a higher degree of heterogeneity.