Permeability Modeling Using Neural Network Approach for Complex Mauddud-Burgan Carbonate Reservoir

Abstract

Abstract This paper reports a comprehensive study towards quantitative characterization of the permeability distribution in complex Mauddud-Burgan reservoir in Kuwait. The main objective in this study is to develop a generalized strategy for data-mining a large data-set of rock measurements. This study utilizes measurements of petrophysical and grain/ pore morphology properties in order to correlate permeability. Data-set contain measurements obtained from different length scales, ranging from SEM to wire-line log scale. Characterizing the permeability for the Mauddud-Burgan reservoir is a challenge because of the complexity of this reservoir. The process is dependant on the type of data available for the reservoir. This study strives toward comprehensive data mining to understand the permeability of this complex reservoir. A Multiple-Layer Feed Forward, MLFF, with back propagation neural network is developed to calculate the permeability at each desired vertical depth in the reservoir. This tool can assist in determining permeability at any vertical depth of the reservoir, within the boundaries of the reservoir model. Knowledge of other petrophysical properties, such as, porosity, pore type, and pore size distribution, as available, are integrated to estimate the permeability. Introduction Historically, in North Kuwait, Mauddud is known to exist as a thick and fractured limestone formation that shows good production. However, in southeast Kuwait at Burgan field, production is difficult. In Burgan field, Mauddud is about 40 feet of net-pay of oil. It shows an average high porosity - about 18% - and a low matrix permeability between 0.8 - 10 millidarcies (1). Understanding the permeability for the Mauddud-Burgan reservoir is important before deciding the method of optimizing the recovery. Understanding this production mechanism is dependant on knowing exactly what is happening to the permeability in the well. Therefore, comprehensive reservoir knowledge is needed to understand the permeability of this complex reservoir. This study strives toward such knowledge. Reservoir modeling through exploiting the reservoir permeability will be the main objective of this study. One of the greatest challenges in this study is securing the data for the analyses. Kuwait Oil Company, KOC, and the Kuwait Ministry of Energy (MOE) reserve the right to access most of the data. Mauddud-Burgan is considered one of the classified reservoirs in Kuwait; therefore, most of the locations and the names of the well contributing to this study are going to be anonymous. The reason for such act, in large, is because of the limitation of availability of sample access and the nature of classified information shared by KOC and MOE. Some data were already processed and available for examination such as, logs. Others needed to be processed such as core analysis, thinsection, and SEM (2). Fortunately, Well #A has the most useful information in regards of the source of the samples under examination. There is missing information in this well, which the model formulated for this study will help to predict and identify this missing information. Well #A is the recent well drilled for Mauddud-Burgan studies, and has the least missing information for that particular reservoir. Well #A is a vertical well. Well #A is considered a classified well in regards to information, which means well name and location will be hidden and referred to in the following data and discussion as symbols such as well #A.