System and method for implementing machine learning for 3D geo-modeling of petroleum reservoirs

1 . A computer-implemented method, comprising:

accessing measurement data that characterize one or more petrophysical features at a range of depths inside a reservoir, wherein the measurement data are from more than one well locations at the reservoir;

generating a filtered measurement dataset, comprising:

detecting anomalous measurement data points by computing, for each measurement data point, a likelihood value using a probability density function based on a mean and a covariance matrix of the one or more petrophysical features, and identifying measurement data points having a likelihood value below a pre-determined threshold; and

automatically removing the anomalous measurement data points;

identifying a plurality of layers along the range of depths of the reservoir based on the filtered measurement dataset;

within each layer of the plurality of layers, assigning, using a machine learning algorithm, filtered measurement data points corresponding to the layer among a plurality of clusters that are non-overlapping, wherein the clusters are determined based on the one or more petrophysical features of the measurement data,

each cluster corresponding to a flow unit that identifies a volume of the reservoir based on one or more of flow potential and rock quality;

generating a three-dimensional (3D) permeability model of the reservoir based on the flow units of each layer from the plurality of layers, wherein the 3D permeability model comprises grid blocks assigned petrophysical feature values, with flow units allocated to the grid blocks based on the assigned petrophysical feature values; and

simulating a performance of the reservoir based on the 3D permeability model.

2 . The computer-implemented method of claim 1 , wherein the one or more petrophysical features comprise: a permeability, a porosity, a water saturation, a rock-type, and a vertical heterogeneity.

3 . The computer-implemented method of claim 1 , further comprising:

comparing the 3D permeability model with internal architectures derived from mercury injection capillary pressure (MICP) experiments on core samples extracted from the more than one well locations; and

adjusting the plurality of layers and the plurality of clusters such that a difference between the 3D permeability model and the internal architectures is reduced.

4 . The computer-implemented method of claim 1 , wherein the machine learning algorithm determines the plurality of clusters, each having a centroid, and wherein a distance from the measurement data to the centroid of each cluster is substantially reduced.

5 . The computer-implemented method of claim 4 , wherein the machine learning algorithm comprises: a K-means clustering algorithm.

6 . The computer-implemented method of claim 1 , further comprising:

determining a number for the plurality of clusters based on overlaying available data from a special core analysis (SCA) on the measurements data.

7 . The computer-implemented method of claim 6 , wherein the SCA generates at least one of: a relative permeability, or a capillary pressure.

8 . The computer-implemented method of claim 1 , further comprising:

associating the measured data grouped among the plurality of clusters with the more than one well locations such that the 3D permeability model covers the more than one well locations.

9 . The computer-implemented method of claim 1 , further comprising:

determining a water saturation height function for each cluster within each layer based on the machine learning algorithm; and

incorporating the water saturation height function for each cluster into the 3D permeability model of the reservoir.

10 . A computer system comprising one or more processors configured to perform operations of:

accessing measurement data that characterize one or more petrophysical features at a range of depths inside a reservoir, wherein the measurement data are from more than one well locations at the reservoir;

generating a filtered measurement dataset, comprising:

detecting anomalous measurement data points by computing, for each measurement data point, a likelihood value using a probability density function based on a mean and a covariance matrix of the one or more petrophysical features, and identifying measurement data points having a likelihood value below a pre-determined threshold; and

automatically removing the anomalous measurement data points;

identifying a plurality of layers along the range of depths of the reservoir based on the filtered measurement dataset;

within each layer of the plurality of layers, assigning, using a machine learning algorithm, filtered measurement data points corresponding to the layer among a plurality of clusters that are non-overlapping, wherein the clusters are determined based on the one or more petrophysical features of the measurement data,

each cluster corresponding to a flow unit that identifies a volume of the reservoir based on one or more of flow potential and rock quality;

generating a three-dimensional (3D) permeability model of the reservoir based on the flow units of each layer from the plurality of layers, wherein the 3D permeability model comprises grid blocks assigned petrophysical feature values, with flow units allocated to the grid blocks based on the assigned petrophysical feature values; and

simulating a performance of the reservoir based on the 3D permeability model.

11 . The computer system of claim 10 , wherein the one or more petrophysical features comprise: a permeability, a porosity, a water saturation, a rock-type, and a vertical heterogeneity.

12 . The computer system of claim 10 , wherein the operations further comprise:

comparing the 3D permeability model with internal architectures derived from mercury injection capillary pressure (MICP) experiments on core samples extracted from the more than one well locations; and

adjusting the plurality of layers and the plurality of clusters such that a difference between the 3D permeability model and the internal architectures is reduced.

13 . The computer system of claim 10 , wherein the machine learning algorithm determines the plurality of clusters, each having a centroid, and wherein a distance from the measurement data to the centroid of each cluster is substantially reduced.

14 . The computer system of claim 13 , wherein the machine learning algorithm comprises: a K-means clustering algorithm.

15 . The computer system of claim 10 , further comprising:

determining a number for the plurality of clusters based on overlaying available data from a special core analysis (SCA) on the measurements data.

16 . The computer system of claim 15 , wherein the SCA generates at least one of: a relative permeability, or a capillary pressure.

17 . The computer system of claim 10 , wherein the operations further comprise:

associating the measured data grouped among the plurality of clusters with the more than one well locations such that the 3D permeability model covers the more than one well locations.

18 . The computer system of claim 10 , wherein the operations further comprise:

determining a water saturation height function for each cluster within each layer based on the machine learning algorithm; and

incorporating the water saturation height function for each cluster into the 3D permeability model of the reservoir.