The classification of lithofacies and their accurate representation in a 3D cellular geologic model is critical to understanding the field characteristics because permeability and fluid saturations for a given porosity and elevation above free water vary considerably among lithofacies. The best source of lithofacies information is the reservoir rock core samples from wells; however, cores are not commonly taken due to excessive expense. Since the availability of core samples is limited compared to the number of wells in the field, developing a method for estimating lithofacies in wells without cores is necessary. In this study, core lithofacies are extrapolated from cored wells (training wells) to uncored wells through the comparison of physical rock properties measured by wire-line logs. Associating well log data with lithofacies can be difficult due to the heterogeneous nature of rocks, especially carbonate rocks. Lithofacies can be defined using any set of rock properties; however, only lithofacies defined by variations in properties that affect well log response can be identified using well log data. Moreover, some useful rock properties, such as porosity and permeability, affect well log response. Artificial Neural Networks (ANNs) are computational models inspired by brain structure mechanisms and functions. The system employs a set of nonlinear and linear activation functions that do not require a priori selection of a mathematical model. Recent applications of ANN to geological studies have demonstrated its effectiveness in prediction, estimation and characterization. Neural networks have been developed and utilized for the solution of a variety of pattern recognition, classification, and signal identification and prediction problems. Neural networks utilized for the prediction of parameters and pattern classification are trained based on data-sets that contain a number of training patterns. Each training pattern is presented as a pair of two components: the input data and the respective classification outcome. After the neural network is trained, it may be applied to a new data-set which contains only the values of input parameters. The goal of this study is to establish a method for lithofacies classification utilizing well data and taking advantage of artificial neural network. Additionally, sonic log efficiency in improving of prediction of reservoir lithofacies is discussed.
In this paper, the main goal is the classification of hydrocarbon reservoir lithofacies, applying an artificial neural network technique, specially back propagation (BP) and Levenberg- Marquwardt training algorithms, on gamma-ray , density, neutron, sonic, and photo electric effect (PEF) logs. Moreover, in this research, the efficiency of sonic log data in lithofacies estimation has been investigated. Paying attention to the fact that lithofacies determination from core experiments can be costly, in this approach, lithofacies identification expenses are reduced by eliminating the need to perform a coring operation.
Data from four wellbores in a hydrocarbon reservoir was used. The network was trained primarily on one of the wells in which the core analysis was available and was subsequently tested in another well that did not play any role in the training phase. After achieving efficiency reliability, the network was applied in lithofacies estimation in two other wells (A1 and A2). The amount of MSE (mean squares error) in this method using only gamma-ray, density, neutron and photoelectric effect logs was 0.068 and 0.074 for wells A1 for A2, respectively.
In the case of using a sonic log in addition to previous inputs, the MSE decreases to 0.052 for well A1 and 0.060 for well A2, which implies an estimation improvement.