Subsurface knowledge enhancement using data science-constrained inverse modelling

Abstract

A method for updating a model of a subsurface formation includes receiving input data for a subsurface formation. The method also includes creating an algorithm based upon the input data. The method also includes determining a plurality of 3D subsurface properties that are possible using the algorithm. The method also includes predicting one or more flowing characteristics of the subsurface formation based upon the 3D subsurface properties. The method also includes training a machine-learning (ML) model using the one or more predicted flowing characteristics.

What is claimed is: 
     
         1 . A method for updating a model of a subsurface formation, the method comprising:
 receiving input data for a subsurface formation;   creating an algorithm based upon the input data;   determining a plurality of 3D subsurface properties that are possible using the algorithm;   predicting one or more flowing characteristics of the subsurface formation based upon the 3D subsurface properties; and   training a machine-learning (ML) model using the one or more predicted flowing characteristics.   
     
     
         2 . The method of  claim 1 , wherein the input data comprises geological features, seismic data, seismic attributes, well logs, structural geometry, rock and fluid physics data, and/or initial conditions. 
     
     
         3 . The method of  claim 1 , wherein the 3D subsurface properties comprise porosity, permeability, and/or initial water saturation. 
     
     
         4 . The method of  claim 1 , wherein the 3D subsurface properties are a function of one or more intervention variables, and wherein the one or more intervention variables are selected to intervene to modify a sampling from the algorithm. 
     
     
         5 . The method of  claim 4 , further comprising adjusting the one or more intervention variables using the trained ML model to provide updated 3D subsurface properties. 
     
     
         6 . The method of  claim 5 , further comprising receiving one or more measured flowing characteristics, wherein the ML model replaces reservoir simulation to expedite determining an objective function that measures a mismatch between the one or more predicted flowing characteristics versus the one or more measured flowing characteristics, and wherein the one or more intervention variables are adjusted to minimize the mismatch. 
     
     
         7 . The method of  claim 5 , wherein adjusting the one or more intervention variables adjusts how the algorithm is sampled, and wherein the algorithm comprises a probabilistic random forest. 
     
     
         8 . The method of  claim 1 , further comprising displaying an output of the trained ML model. 
     
     
         9 . The method of  claim 1 , wherein the one or more predicted flowing characteristics comprise pressure and/or flow rates at one or more wells. 
     
     
         10 . The method of  claim 9 , further comprising performing a wellsite action in response to an output of the trained ML model, wherein the wellsite action adjusts the pressure and/or the flow rate in the one or more wells. 
     
     
         11 . A computing system, comprising:
 one or more processors; and   a memory system comprising one or more non-transitory computer-readable media storing instructions that, when executed by at least one of the one or more processors, cause the computing system to perform operations, the operations comprising:
 receiving input data for a subsurface formation, wherein the input data comprises geological features, seismic data, seismic attributes, well logs, structural geometry, rock and fluid physics data, and/or initial conditions; 
 creating a probabilistic random forest based upon the input data; 
 determining a plurality of 3D subsurface properties that are possible using the probabilistic random forest, wherein the 3D subsurface properties comprise porosity, permeability, and/or initial water saturation, wherein the 3D subsurface properties are a function of one or more intervention variables, wherein the one or more intervention variables are selected to intervene to modify a sampling from the probabilistic random forest; 
 predicting one or more flowing characteristics of the subsurface formation based upon the 3D subsurface properties, wherein the one or more predicted flowing characteristics comprise pressure and/or flow rates at one or more wells; 
 training a machine-learning (ML) model using the one or more predicted flowing characteristics; 
 receiving one or more measured flowing characteristics, wherein the one or more measured flowing characteristics comprise measured pressure and measured flow rates at the one or more wells; and 
 adjusting the one or more intervention variables using the trained ML model to provide updated 3D subsurface properties, wherein adjusting the one or more intervention variables adjusts how the probabilistic random forest is sampled. 
   
     
     
         12 . The computing system of  claim 11 , wherein the one or more intervention variables comprise external variables and/or secondary input data that are selected to intervene to modify the sampling from the probabilistic random forest, wherein the external variables are not used to train the probabilistic random forest, and wherein the secondary input data are used to train the probabilistic random forest. 
     
     
         13 . The computing system of  claim 11 , wherein the one or more flowing characteristics are predicted using a physics simulator, and wherein the trained ML model functions as a surrogate model for the physics simulator. 
     
     
         14 . The computing system of  claim 11 , wherein the trained ML model comprises a physics-informed neural operator, a Fourier neural operator, a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory (LSTM), a physics-informed neural network (PINN), a graph neural network (GNN), or other ML-based surrogate model. 
     
     
         15 . The computing system of  claim 11 , wherein the one or more intervention variables are adjusted by an optimizer to minimize a mismatch between the one or more predicted flowing characteristics versus the one or more measured flowing characteristics. 
     
     
         16 . A non-transitory computer-readable medium storing instructions that, when executed by one or more processors of a computing system, cause the computing system to perform operations, the operations comprising:
 receiving input data for a subsurface formation, wherein the input data comprises geological features, seismic data, seismic attributes, well logs, structural geometry, rock and fluid physics data, and/or initial conditions;   creating a probabilistic random forest based upon the input data;   determining a plurality of 3D subsurface properties that are possible using the probabilistic random forest, wherein the 3D subsurface properties comprise porosity, permeability, and/or initial water saturation, wherein the 3D subsurface properties are a function of one or more intervention variables, wherein the one or more intervention variables comprise external variables and/or secondary input data that are selected to intervene to modify a sampling from the probabilistic random forest, wherein the external variables are not used to train the probabilistic random forest, and wherein the secondary input data are used to train the probabilistic random forest;   predicting one or more flowing characteristics of the subsurface formation based upon the 3D subsurface properties, wherein the one or more flowing characteristics are predicted using a physics simulator, and wherein the one or more predicted flowing characteristics comprise pressure and/or flow rates at one or more wells;   training a machine-learning (ML) model using the one or more predicted flowing characteristics, wherein the trained ML model functions as a surrogate model for the physics simulator, and wherein the trained ML model comprises a physics-informed neural operator, a Fourier neural operator, or any other model capable of functioning as the surrogate model;   receiving one or more measured flowing characteristics, wherein the one or more measured flowing characteristics comprise measured pressure and measured flow rates at the one or more wells; and   adjusting the one or more intervention variables using the trained ML model to provide updated 3D subsurface properties, wherein adjusting the one or more intervention variables adjusts how the probabilistic random forest is sampled, and wherein the one or more intervention variables are selected by an optimizer to minimize a mismatch between the one or more predicted flowing characteristics versus the one or more measured flowing characteristics.   
     
     
         17 . The non-transitory computer-readable medium of  claim 16 , wherein the operations further comprise displaying the updated 3D subsurface properties. 
     
     
         18 . The non-transitory computer-readable medium of  claim 16 , wherein the operations further comprise performing a wellsite action based upon the updated 3D subsurface properties. 
     
     
         19 . The non-transitory computer-readable medium of  claim 18 , wherein the wellsite action comprises generating and/or transmitting a signal that recommends, instructs, or causes a physical action to occur in or to the one or more wells. 
     
     
         20 . The non-transitory computer-readable medium of  claim 19 , wherein the physical action comprises adjusting the pressure in the one or more wells using a pump at the surface, adjusting the flow rates into and/or out of the one or more wells using the pump, selecting where to drill a new well, drilling the new well, varying a weight and/or torque on a drill bit that is drilling the new well, varying a drilling trajectory of the new well, or varying a concentration and/or flow rate of a fluid pumped into the new well.