About This Item

Share This Item

The AAPG/Datapages Combined Publications Database

GCAGS Transactions

Abstract


Gulf Coast Association of Geological Societies Transactions
Vol. 49 (1999), Pages 20-22

EXTENDED ABSTRACT: Classification of Porosity and Permeability Category with Regression Trees

Haiyan Li (1), Hui-Chuan Chen (1) and Ernest Mancini (2)

The University of Alabama, Tuscaloosa, AL.

ABSTRACT

Permeability and porosity are two of the most important petrophysical parameters in reservoir characterization. In a carbonate reservoir, the exact porosity/permeability values are often too difficult to be obtained from well-log data. Nevertheless, knowledge of relative porosity/permeability category (high, moderate or low categories) is useful for formulating strategies for oil and gas field development and production. This paper presents a classification method capable of determining the porosity/permeability category by feet from well-log data.

Regression trees, which incorporate regression method, tree structure, and expert knowledge to determine data category, are utilized. A regression tree can be constructed based on a a training well. This well contains the independent variables obtained from well logs and the dependent variables (in terms of desired categories) obtained from core data. The constructed tree is then used to explore the predictive structure of the given data and the interactions between various log signatures. Finally a classification method is used to derive the corresponding category of porosity/permeability categories.

Data from seven wells in North Choctaw Ridge Smackover Formation, located in Choctaw County, AL are used in this study. From each well, seven well-log signatures were selected as independent variables for classification. They are well-log depth, gamma ray, density porosity, neutron porosity, lateral induction, medium induction and dual induction. The corresponding dependent variables are porosity and permeability form core analysis. The proposed classifier was trained on data from well 1944, and then tested on wells 1688, 1712, 1739, 1835, 1878 and 1946. To measure the prediction accuracy of the classifier, the output of these six test wells was compared with actual core porosity and core permeability. Table 1 lists the error percentage of the predicated porosity of reach of the seven wells. Table 2 lists the error percentage of the predicted permeability. From Tables 1 and 2, the proposed classifier yields a 86.7% accuracy for porosity and 82.8% accuracy for permeability. Figure 1 compares the predicated porosity with the core-measured porosity by feet for each of the seven wells.

Table 1. Error rate for porosity prediction

Table 2. Error rate for permeability prediction

Using the same data, a backpropagation neural network with three layers (input, hidden, and output) was also applied. As a result, a 73.2% accuracy and 62.2% accuracy were obtained for porosity and permeability, respectively. The proposed classification method yields higher accuracy than the popular backpropagation neural network in the classification of porosity and permeability category.

End_Page 20------------------------

Figure 1.

End_Page 21------------------------

Figure 2.

End_of_Record - Last_Page 22-------

Pay-Per-View Purchase Options

The article is available through a document delivery service. Explain these Purchase Options.

Watermarked PDF Document: $14
Open PDF Document: $24