Distality management in automatic well correlation.

G. Vix and Jonathan Edwards and Guillaume Caumon. ( 2015 )
in: 35th Gocad Meeting - 2015 RING Meeting, ASGA

Abstract

Several well correlation paradigms can be used to obtain a consistent identification of heterogeneities and stratigraphic units, both at reservoir and basin scales. Depending on the correlation strategy and on the available data, uncertainty exists and may affect the forecasted reservoir behavior. Well correlation is usually done manually, but it can be interesting to use automatic methods to reduce subjectivity and assess correlation uncertainties. The advantage of an automatic and stochastic method is to quickly build lots of possible models of reservoir layering, and also to decrease the impact of the interpreter’s undocumented choices. In this paper, we use a stochastic well correlation method derived from the Dynamic Time Warping (DTW) algorithm, which matches two signals based on the sum of their local similarities. The input of the method is primarily a set of stratigraphic sequences identified at two wells. We use, as in many manual correlations, the principles of sequence stratigraphy to define correlation rules used by the DTW method. The similarity between two given sequences on two wells may be defined on their type (transgressive or regressive) and their thicknesses for example. It is then possible to minimize lateral thickness variations during automatic correlation. This means that correlations between sequences with the same thickness are favored. However, this is not always observed on the field. For instance, a sequence in regression tends to have decreasing thickness from the proximal pole to the distal pole (i.e., from the upstream to the downstream of the sediment transport direction). Conversely, a sequence in regression thickens in the same direction. We propose to account for this observation by taking as input a distality vector which links the proximal pole to the distal pole. This type of vector can typically be defined from basin regional knowledge. In order to reproduce this phenomenon, we propose a correlation cost function which: vertically sets if the sequences should become thicker or thinner, and manages the correlation costs accordingly; horizontally evaluates the impact of the distality on the correlation. Indeed, if the vector extracted from the wells positions on surface and the distality vector are parallel, the impact of the distality should be maximum. However, if those two vectors are perpendicular, the distality should have no impact on the layer thickness. We implemented this method in a plugin of a geomodeling software and applied it to synthetic data. The obtained results significantly improve the obtained correlations likelihood as compared to simpler rules neglecting the distality. Overall, the results match the expected results: the sequences thicken and slim down according to the distality vector and type of sequences. This work opens interesting perspectives to better capture geological uncertainties in reservoir and basin models, and understand their impact on hydrocarbon migration during basin evolution and reservoir recovery.

Download / Links

BibTeX Reference

@INPROCEEDINGS{VixGM2015,
    author = { Vix, G. and Edwards, Jonathan and Caumon, Guillaume },
     title = { Distality management in automatic well correlation. },
 booktitle = { 35th Gocad Meeting - 2015 RING Meeting },
      year = { 2015 },
 publisher = { ASGA },
  abstract = { Several well correlation paradigms can be used to obtain a consistent identification of heterogeneities and stratigraphic units, both at reservoir and basin scales. Depending on the correlation strategy and on the available data, uncertainty exists and may affect the forecasted reservoir behavior. Well correlation is usually done manually, but it can be interesting to use automatic methods to reduce subjectivity and assess correlation uncertainties. The advantage of an automatic and stochastic method is to quickly build lots of possible models of reservoir layering, and also to decrease the impact of the interpreter’s undocumented choices. In this paper, we use a stochastic well correlation method derived from the Dynamic Time Warping (DTW) algorithm, which matches two signals based on the sum of their local similarities. The input of the method is primarily a set of stratigraphic sequences identified at two wells. We use, as in many manual correlations, the principles of sequence stratigraphy to define correlation rules used by the DTW method. The similarity between two given sequences on two wells may be defined on their type (transgressive or regressive) and their thicknesses for example. It is then possible to minimize lateral thickness variations during automatic correlation. This means that correlations between sequences with the same thickness are favored. However, this is not always observed on the field. For instance, a sequence in regression tends to have decreasing thickness from the proximal pole to the distal pole (i.e., from the upstream to the downstream of the sediment transport direction). Conversely, a sequence in regression thickens in the same direction. We propose to account for this observation by taking as input a distality vector which links the proximal pole to the distal pole. This type of vector can typically be defined from basin regional knowledge. In order to reproduce this phenomenon, we propose a correlation cost function which: vertically sets if the sequences should become thicker or thinner, and manages the correlation costs accordingly; horizontally evaluates the impact of the distality on the correlation. Indeed, if the vector extracted from the wells positions on surface and the distality vector are parallel, the impact of the distality should be maximum. However, if those two vectors are perpendicular, the distality should have no impact on the layer thickness. We implemented this method in a plugin of a geomodeling software and applied it to synthetic data. The obtained results significantly improve the obtained correlations likelihood as compared to simpler rules neglecting the distality. Overall, the results match the expected results: the sequences thicken and slim down according to the distality vector and type of sequences. This work opens interesting perspectives to better capture geological uncertainties in reservoir and basin models, and understand their impact on hydrocarbon migration during basin evolution and reservoir recovery. }
}