Automatic Identification of Calcareous Lithologies Using Support Vector Machines, Borehole Logs and Fractal Dimension of Borehole Electrical Imaging

Earth sciences research journal

Ochoa G., Luis H.¹  Leal F., Jorge A.² 

[1] Schlumberger Technology, U.S;Universidad Nacional de Colombia, Bogotá, Colombia

关键词: machine learning;    support vector machines;    borehole logs;    image logs;    fractal dimension;    calcareous lithologies;    Catatumbo Basin.;   

DOI  :  10.15446/esrj.v22n2.68320

学科分类：天文学(综合)

来源: Universidad Nacional de Colombia * Departamento de Geociencias

【 摘 要 】

In this research two support vector machines (SVMs) and a logic function were applied to identify calcareous sections automatically in wells located in the former Barco Concession, Catatumbo Basin - Colombia. During training stages the SVMs used nuclear logs, such as neutron, photoelectric factor and gamma ray in order to differentiate calcareous from clastic sections; additionally, in this stage the fractal dimension of resistive images along with mean and variance of resistivity acquired with imaging tool (of high resolution) are employed to identify textural features of the rocks. The first SVM also employed in the training stage intervals manually interpreted of fossiliferous limestone, performed by a specialized geologist integrating core and logs information of a pilot well; during classification stage, this SVM automatically recognized intervals with fossiliferous limestone by using only logs data of any well of the field. The second SVM was also trained with nuclear logs, resistivity and fractal dimension, but in this case, with information of intervals composed of calcareous shales interbedded with limestone, recognizing automatically these rock associations during classification stage without interpretation of a geologist as input data. Finally, a logic function was applied to intervals with photoelectric factor ≥ 4 and all sections not classified by the SVMs were grouped as laminated calcareous rocks. The SVMs and logic function show accuracy of 98.76 %, 94.02 % and 94.60 % respectively in five evaluated wells and can be applied to other wells in the field that have the same dataset conditions. This methodology is dependent of the data quality and all intervals affected by poor borehole conditions should be removed in order to avoid erratic interpretations. This model must to be recalibrated in case to be applied in other fields of the basin.

【 授权许可】

CC BY