Static engine and neural network for a cognitive reservoir system

Abstract

Implementations described and claimed herein provide systems and methods for developing a reservoir. In one implementation, observed data points in a volume along a well trajectory and well logs corresponding to observed data points are received at a neural network. Feature vectors are generated using the neural network. The feature vectors are defined based on a distance between each of the observed data points and randomly generated points in the volume. A 3D populated log is generated by propagating well log values of the feature vectors across the volume. Uncertainty is quantified by generating a plurality of realizations including the 3D populated log. Each of the realizations is different and equally probable. core values are generated from the realizations, and a static model of the reservoir is generated by clustering the volume into one or more clusters of rock types based on the core values.

1 . A method for developing a reservoir, the method comprising:
 receiving static data at a static modeler, the static data including a set of observed data points in a reservoir volume along a well trajectory and well logs corresponding to the set of observed data points;   generating a set of feature vectors to learn neural networks, the set of feature vectors defined based on a distance between at least one of the observed data points and a set of randomly selected points in the reservoir volume;   generating a three-dimensional populated log by propagating well log values of the set of feature vectors across the reservoir volume;   quantifying uncertainty by applying the learned neural networks to generate a plurality of realizations including the three-dimensional populated log;   changing, by the static modeler, the set of randomly selected points, wherein each of the plurality of realizations is different and equally probable; and   generating a plurality of static models comprising a random sample of three-dimensional populated log data of the plurality of realizations.   
     
     
         2 . The method of  claim 1 , the method further comprising:
 clustering the reservoir volume into a plurality of clustered three-dimensional volumes using a learned clustering algorithm alongside randomly selected three-dimensional log data from the plurality of realizations, wherein the static models include one of the clustered three-dimensional volumes.   
     
     
         3 . The method of  claim 2 , wherein at least one of the static models includes one or more fault planes fitted to the plurality of clustered three-dimensional volumes. 
     
     
         4 . The method of  claim 2 , wherein a reservoir graph is constructed from at least one of the static models with at least one three-dimensional volumes represented as a vertex. 
     
     
         5 . The method of  claim 4 , wherein the reservoir graph is defined such that at least one of the clustered three-dimensional volumes has at least one of: at most one pressure observation point, only one well passing through the cluster, or a spatially continuous voxel set only. 
     
     
         6 . The method of  claim 1 , the method further comprising:
 populating core data in the reservoir volume using a learned k-nearest neighbor algorithm alongside the three-dimensional populated log, wherein the static models include a three-dimensional volume of the core data, and wherein the core data includes at least one of porosity, permeability, or water saturation.   
     
     
         7 . The method of  claim 1 , wherein the three-dimensional populated log data includes at least one of gamma ray, neutron porosity, bulk density, or resistivity. 
     
     
         8 . One or more non-transitory tangible computer-readable storage media storing computer-executable instructions for performing a computer process on a computing system, the computer process comprising:
 receiving static data at a static modeler, the static data including a set of observed data points in a reservoir volume along a well trajectory and well logs corresponding to the set of observed data points;   generating a set of feature vectors to learn neural networks, the set of feature vectors defined based on a distance between each of the observed data points and a set of randomly selected points in the reservoir volume;   generating a three-dimensional populated log by propagating well log values of the set of feature vectors across the reservoir volume;   quantifying uncertainty by applying the learned neural networks to generate a plurality of realizations including the three-dimensional populated log;   changing, by the static modeler, the set of randomly selected points, wherein each of the plurality of realizations is different and equally probable; and   generating a plurality of static models comprising a random sample of three-dimensional populated log data of the plurality of realizations.   
     
     
         9 . The one or more non-transitory tangible computer-readable storage media of  claim 8 , wherein the computer process further comprising:
 clustering the reservoir volume into a plurality of clustered three-dimensional volumes using a learned clustering algorithm alongside randomly selected three-dimensional log data from the plurality of realizations, wherein the static models include one of the clustered three-dimensional volumes.   
     
     
         10 . The one or more non-transitory tangible computer-readable storage media of  claim 9 , wherein at least one of the static models includes one or more fault planes fitted to the plurality of clustered three-dimensional volumes. 
     
     
         11 . The one or more non-transitory tangible computer-readable storage media of  claim 10 , wherein a reservoir graph is constructed from the static model with at least one of the one or more clusters represented as a vertex. 
     
     
         12 . The one or more non-transitory tangible computer-readable storage media of  claim 11 , wherein the reservoir graph is defined such that at least one of the clusters has at least one of: at most one pressure observation point, only one well passing through the cluster; or a spatially continuous voxel set only. 
     
     
         13 . The one or more non-transitory tangible computer-readable storage media of  claim 8 , wherein the computer process further comprising:
 populating core data in the reservoir volume using a learned k-nearest neighbor algorithm alongside the three-dimensional populated log, wherein the static models include a three-dimensional volume of the core data, and wherein the core data includes at least one of porosity, permeability, or water saturation.   
     
     
         14 . The one or more non-transitory tangible computer-readable storage media of  claim 8 , wherein the three-dimensional populated log data includes at least one of gamma ray, neutron porosity, bulk density, or resistivity. 
     
     
         15 . A system for developing a reservoir, the system comprising:
 one or more processors; and   at least one non-transitory computer readable medium having stored therein instructions executed by the one or more processors to:
 receive static data at a static modeler, the static data including a set of observed data points in a reservoir volume along a well trajectory and well logs corresponding to the set of observed data points; 
 generate a set of feature vectors to learn neural networks, the set of feature vectors defined based on a distance between each of the observed data points and a set of randomly selected points in the reservoir volume; 
 generate a three-dimensional populated log by propagating well log values of the set of feature vectors across the reservoir volume; 
 quantify uncertainty by applying the learned neural networks to generate a plurality of realizations including the three-dimensional populated log; 
 change, by the static modeler, the set of randomly selected points, wherein each of the plurality of realizations is different and equally probable; and 
 generate a plurality of static models comprising a random sample of three-dimensional populated log data of the plurality of realizations. 
   
     
     
         16 . The system of  claim 15 , wherein the instructions further execute the one or more processor to:
 clustering the reservoir volume into a plurality of clustered three-dimensional volumes using a learned clustering algorithm alongside randomly selected three-dimensional log data from the plurality of realizations, wherein the static models include one of the clustered three-dimensional volumes.   
     
     
         17 . The system of  claim 16 , wherein at least one of the static models includes one or more fault planes fitted to the plurality of clustered three-dimensional volumes. 
     
     
         18 . The system of  claim 17 , wherein a reservoir graph is constructed from the static model with each of the one or more clusters represented as a vertex. 
     
     
         19 . The system of  claim 18 , wherein the reservoir graph is defined such that at least one of the clustered three-dimensional volumes has at least one of: at most one pressure observation point, only one well passing through the cluster, or a spatially continuous voxel set only. 
     
     
         20 . The system of  claim 15 , wherein the instructions further execute the one or more processor to:
 populating core data in the reservoir volume using a learned k-nearest neighbor algorithm alongside the three-dimensional populated log, wherein the static models include a three-dimensional volume of the core data, and wherein the core data includes at least one of porosity, permeability, or water saturation.