Well logs reconstruction by DCT based IPRM and genetic algorithm

• Published in: Journal of petroleum science & engineering Vol. 195; p. 107846
• Main Authors: Ahmadi, Kamyar, Riahi, Mohammad Ali, Anzehaee, M. Mousavi
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2020
• Subjects: Correlation; Discrete cosine transform; Gegenbauer polynomials; Genetic algorithm; Gibbs phenomenon; Inverse polynomial reconstruction; Perforation; Self-organizing map; Correlation; Gegenbauer polynomials; Gibbs phenomenon; Inverse polynomial reconstruction; Perforation; Self-organizing map; Genetic algorithm; Discrete cosine transform
• ISSN: 0920-4105; 1873-4715
• DOI: 10.1016/j.petrol.2020.107846

Hydrocarbon production from a particular layer requires careful identification of the stratigraphic boundaries and choosing the right perforation against the face of oil-bearing intervals. Excessive perforations may cause unwanted water production from adjacent layers. Nowadays, clustering of well log data through various methods such as self-organized maps (SOM) helps to find hydrocarbon intervals from other layers. SOM is a type of artificial neural network, which uses the non-monitored learning of the multi-dimensional logging observations to create a 2D clustering map. However, the noisy response data greatly reduces accuracy of the methods by increasing correlation of different data groups. Although spectral such as discrete wavelet transform (DWT) can reduce correlation and reconstruct the logs that no longer fluctuate as before, they have two major limitations. First, responses smooth the edge-shape behaviors in the log curve while preserving them is essential for recognizing the stratification changes. The second limitation is the generation of spurious oscillations in the reconstructed signals at the vicinity of the edges, known as the Gibbs phenomena. At the present study to overcome these problems, we use, for the first time, the inverse polynomial reconstruction method (IPRM) based on discrete cosine transform (DCT) to attenuate the random noise by projecting the data first on the orthogonal bases derived from Gegenbauer polynomials and then on cosine space. The de-noising level depends on the minimum description length (MDL) criterion that minimizes the distance of all the corresponding coefficients between the original and reconstructed signals during 100 iterations. However, in the proposed method, we use a genetic algorithm to decide quickly and optimally the coefficients of DCT and Gegenbauer required in real data reconstruction related to one of the oil fields in southwestern Iran. The results show that this preprocessing procedure compared to DCT and DWT, increases the success rate of the SOM method for oil-bearing zone detection based on evidence such as core drilling and perforation data. At the same time, the computational time is much shorter than the usual IPRM. The optimal perforation subsequent to this study may cut unwanted water production from adjacent layers. Such water with radioactive nature, in addition to causing environmental problems, greatly increases the cost of extraction operations due to the need to install surface-separating equipment. •DCT-based IPRM increases the success rate of the SOM modeling in terms of matching core and production data of the oil field.•This pre-processing procedure specifically weakened the noise of the input logs to SOM modeling.•Unlike conventional spectral methods, IPRM preserves the edge-like behavior of well log curve supporting the boundaries.•The genetic algorithm decides the ideal number of DCT and Gegenbauer coefficients required in IPRM.•The proposed methodology, can cut the amount of unwanted water production and find the perforation locations appropriately.