IMPROVING THE ACCURACY OF RESERVOIR PROPERTIES PREDICTION USING MACHINE LEARNING METHODS
E.I. Korytkin1,2, G.M. Mitrofanov1,3,4
1Trofimuk Institute of Petroleum Geology and Geophysics, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia
2OOO SakhalinNIPI Nefti i Gaza, Yuzhno-Sakhalinsk, Russia
3Novosibirsk State University, Novosibirsk, Russia
4Novosibirsk State Technical University, Novosibirsk, Russia
Keywords: 3D seismic exploration, classification, Bayesian classifier, prior probabilities, seismic facies extraction
Abstract
The article considers the issues of determining the characteristics of target horizons using methods capable of learning on large volumes of heterogeneous data and high prediction accuracy. The methods are used to solve problems of seismic facies analysis at oil and gas fields, the main purpose of which is to reconstruct the sedimentatry rocks and predict lithofacies in the study area. The object of the study was one of the fields in the Volga-Ural region. An improved Bayesian classifier was used as a tool. It was used to determine promising distribution zones of the productive B2 formation reservoir of the Bobrikovian deposits of the Lower Carboniferous and to assess the hydrocarbon production potential. During the research, the effectiveness of the application of machine learning methods and the proposed improvements was analyzed.
|
