Permeability and E_ective Porosity wireline logs estimation with over 90% ac-curacy.
•
Use of Borehole Image Data Logs (Ultrasonic and Microresistivity) as only source of information for the petrophysical logs estimation.
•
Derivation of confidence levels for permeability and porosity using of stateof-art techniques in a Bayesian deep learning framework.
•
The correlation between Ultrasonic and Resistivity Image Data Logs and petro-physical logs (Porosity and Permeability) is shown to be objective in this paper, rather than something subjective or intuitive (as it used to be in the past, when petrophysicists could not quantify objectively the amount of information from these data sources that should be taken into account).
Abstract
Image wirelogs, both acoustic and resistivity, encapsulate information regarding the studied reservoir's petrophysical properties that are used to pursue a meaningful characterization of it. More recently, some authors used those image logs to estimate porosity and permeability using a traditional image processing pipeline or using a deep learning perspective. Among others, those quantities might be critical to determine the feasibility of a specific hydrocarbon reservoir exploitation. However, the use of such estimates are limited since it is not possible to assign uncertainty on those deterministic methods. Thus, their reliability can not be assessed. In this work, we propose a novel approach using Deep Learning and Bayesian methodologies to address the estimation of uncertainty quantification for both permeability and porosity simultaneously from the image logs. The current methodology can be used to derive probability density functions (PDFs) from each prediction. As a result, the PDFs were used to estimate the robustness and uncertainties of the predicted values. The current method achieved a correlation coefficient higher than in both porosity and permeability regression in the blind test sample, i.e., using a different set of data that was not applied in the training procedure.
