Application of Artificial Neural Networks in Identification of Geological Formations on the Basis of Well Logging Data – A Comparison of Computational Environments’ Efficiency

Abstract

The paper presents the application of artificial neural networks in lithology identification on the basis of well logging data. The problem is very important considering petroleum geophysics as it allows to find sweet spots -potential deposits of hydrocarbons (oil and gas). The use of advanced statistical methods such as artificial neural networks is expected to improve geological interpretation of geophysical data. Moreover, such methods are capable of dealing with big data sets since well logging provides more and more information about petrophysical (e.g. porosity, density, resistivity, natural gamma radiation, sonic wave propagation) and chemical rock properties (mineral content and element abundance). Therefore, the analyzed data comprises around 56000 records. Two different computational environments has been used in order to examine their efficiency in terms of accuracy of a lithological classification. Computation was done in R software, which is an open source environment, and STATISTICA v. 13 which is a commercial one. As an input, logging data from three boreholes drilled in the Baltic Basin, North Poland were used. The results show that R offers more possibilities of modification of a net. However, STATISTICA provides more user-friendly interface and better accuracy of lithology identification.