Academic Journals Database
Disseminating quality controlled scientific knowledge

Four Classifiers Used in Data Mining and Knowledge Discovery for Petroleum Exploration and Development

Author(s): Guangren SHI

Journal: Advances in Petroleum Exploration and Development
ISSN 1925-542X

Volume: 2;
Issue: 2;
Start page: 12;
Date: 2011;
Original page

Keywords: Multiple regression analysis | Bayesian discrimination | Back-propagation neural network | Support vector machine | Trap quality evaluation | Oil identification

The application of data mining and knowledge discovery in databases for petroleum exploration and development (PE&D) is becoming promising, though still at an early stage. Up to now, the data mining tools usually used in PE&D are four classifiers: multiple regression analysis (MRA), Bayesian discrimination (BAYD), back-propagation neural network (BPNN), and support vector machine (SVM). Each of the four classifiers has its advantages and disadvantages. A question, however, has been raised in applications is: which classifier is the most applicable to a specified application? This paper has given an answer to the question through two case studies: 1) trap quality evaluation of the Northern Kuqa Depression of the Tarim Basin in western China, and 2) oil identification of the Xiefengqiao anticlinal structure of the Jianghan Basin in central China. Case 1 shows that the results of BAYD, BPNN and SVM are same and can have zero residuals, while MRA has unallowable residuals; but Case 2 shows that the results of only SVM have zero residuals, while BAYD, BPNN and MRA have unallowable residuals. The reasons are: a) since the two cases are nonlinear problems, the linear MRA is not applicable; b) since the nonlinearity of Case 1 is weak, the nonlinear BAYD, BPNN and SVM are applicable; and c) since the nonlinearity of Case 2 is strong, only nonlinear SVM is applicable. Therefore, it is proposed that: we can adopt MRA when a problem is linear; adopt BAYD, BPNN, or SVM when a problem is weakly nonlinear; and adopt only SVM when a problem is strongly nonlinear. In addition, the predictions of the applicable classifiers coincide with real exploration results, and a commercial gas trap was discovered after the forecast in Case 1 and SVM can correct some erroneous well-log interpretations in Case 2.Key words: Multiple regression analysis; Bayesian discrimination; Back-propagation neural network; Support vector machine; Trap quality evaluation; Oil identification
RPA Switzerland

Robotic Process Automation Switzerland


Tango Rapperswil
Tango Rapperswil