Q55 : Intelligent detection of fracture specifications in petro-physic well logs
Thesis > Central Library of Shahrood University > Computer Engineering > MSc > 2014
Authors:
Abstarct: Almost 50 percent of oil reservoirs through the world are naturally fractured and in Iran this ratio is up to 90 percent. While accurate fractures determining of an oil field is a very huge work which needs step by step documentation and descxription of huge number of image well logs by experts. In addition already and especially in Iran there are many obstacles in the way of preparing image well logs, this problem have made it unavoidable using other datasets for detecting fractures.
Until now many different studies have been developed in order of separating between open and close natural fractures but none of them achieved the end goal of usability in the industry. In this study it’s tried to use petrophysical data for finding out if a zone of the well is not fractured, open fractured or close fractured.
Suggested method uses pre-processed petrophysical data instead of raw data for data mining. This pre-processing includes windowing method following with replacing the data with the variance of windowed sub-data-signal. After pre-processing and for data classification, we use two classifiers which finally a majority-voting algorithm ensembles their answers.
We experienced many different classification algorithms and at last Naïve Bayes and Random Tree which showed the most correct classification results have been selected to be used in this regard.
This study has been developed on some oil fields of Gachsaran oil field in south west of Iran. Because of some shortages in the data, we proposed two series of experiments, one of them including the data from 7 wells with 7 well logs and the other one including 4 wells with 11 well logs.
Although the main approach of this study aims to data classification into three classes of open, close and not fractured zones, but also in the last section we proposed our suggested method for classifying the data into two classes of fractured and not-fractured zones, so other researchers can compare their two class classifying results with the suggested method.
Throughout the study one of the main principles is to establishing balance between accuracy of three classes. But unfortunately this is not compatible with the nature of natural fractures, because these three classes have no balance and not-fractured is the dominant class. So we are constrained to accept only well balanced classification results which indeed are not the most accurate results.
In conclusion, final results shows the accuracy of 77.5 percent for two classes and more than 80.6 for three classes classification in the best cases. Also the mean of accuracy for two classes is more than 70.1 and for three classes is more than 68.6 percent.
Keywords:
#Data mining #fracture #open fracture #close fracture #classification #Naïve Bayes #Random Tree #Majority Voting #Petrophysical well-log #Image well-log #Windowing. Link
Keeping place: Central Library of Shahrood University
Visitor:
Keeping place: Central Library of Shahrood University
Visitor: