Services on Demand
Journal
Article
Indicators
- Cited by SciELO
- Access statistics
Related links
- Cited by Google
- Similars in SciELO
- Similars in Google
Share
Boletín de Geología
Print version ISSN 0120-0283On-line version ISSN 2145-8553
Abstract
PARDO-DIAZ, Mauro Felipe and VARGAS-JIMENEZ, Carlos Alberto. Photoelectric factor (PEF) inferences in well logs with machine learning. Bol. geol. [online]. 2021, vol.43, n.1, pp.193-210. Epub Jan 01, 2021. ISSN 0120-0283. https://doi.org/10.18273/revbol.v43n1-2021010.
Conventional well logs are important for performing petrophysical analysis, seismic well ties and stratigraphic correlation. This study proposes a methodology to predict these types of logs using machine learning (ML), a tool highly applied in multiple disciplines. The training software used was WEKA (Waikato Environment for Knowledge Analysis), in which a model for the prediction of the Photoelectric Absorption log (PDPE) was generated, based on three attributes, the Gamma Ray log (GRGC), Density log (DEN) and Density Correction log (DCOR). This methodology was applied to well logs of San Fernando Formation, whose equivalent unit would be Mirador Formation, in the southwestern sector of the Llanos Basin, Colombia. Thirteen wells were used to train the model and six other wells were used to evaluate its performance. The results confirm the possibility of correlating logs that measure different characteristics in the rocks and show that inferences in well logs with ML require a detailed filtering to take the trend of the data, and a clear optimization to prevent overfitting in the model.
Keywords : Graph prediction; Attribute selection; Grid search; WEKA; Overfitting.