AAPG Annual Convention and Exhibition

Datapages, Inc.Print this page

Auto Recognition of Carbonate Sedimentary Facies Based on an Improved BP Neural Network

Abstract

Clustering and identification of carbonate sedimentary facies from well logs are difficult due to the strong diagenetic changes of carbonate reservoir, especially just based on conventional well logs, e.g. gamma, acoustic, and resistivity. Though the traditional methods, e.g. cross-plot, fuzzy clustering and Bayes step discrimination method, could recognize some facies, e.g. lagoon facies, backshoal facies and foreshoal facies, they couldn't distinguish reef facies and shoal facies from other facies well. We proposed in this paper an improved BP neural network with better stability and suitability, optimized by an original algorithm as the classifier to solve the sedimentary facies’ auto discrimination of Mishrif formation from the Rumaila oil field in Iraq. The sedimentary pattern of Mishrif formation in Iraq is a typical porous carbonate ramp, forming during the formation of Mesopotamia passive plate boundary from Early Cenomanian to Late Turonian, which was not destroyed by Zagros orogenic movement. 8 wells with complete core, borehole and log data were chosen as the standard wells and 54 samples were inferred from these 8 wells. In order to find a parameter collection matching with the sedimentary facies samples well, we extracted the average value of gamma, neutron and density logs as well as the sum of squares of deviations of gamma as the key parameters. Log facies (facies from log measurements) – sedimentary facies transforming model was established based on the chosen log parameter collection. The total 54 microfacies samples were divided into 12 kinds of log facies and 6 kinds of sedimentary microfacies, e.g. lagoon facies, reef facies, backshoal facies, shoal facies and foreshoal facies. Using the improved BP neural network method, we built the sedimentary facies log discriminating template just based on the conventional well log data and depicted the sedimentary facies of wells without core data precisely with it. Both the reef facies and shoal facies have the similar low gamma value, high neutron value and low density value because of the conformable diagenesis processes they experienced. As a result, the traditional methods have no ability to differentiate these two facies just based on these logs. Compared with traditional methods, the improved one could integrate more value and shape details of log data, which makes it adapt to the quantitative sedimentary facies recognition much better, with a correct rate higher than 85%.