Estimating productivity and estimated ultimate recovery (eur) of unconventional wells through spatial-performance relationship using machine learning

Abstract

Systems and methods include an importance that each of the attributes and features of the well data has on machine learning models. Well data is collected for each well in an unconventional field, including attributes and features of basin data, completion data, and production data. Spatial features are generated for each well in different regions. A combined well features dataset is generated. The dataset maps the well data to the spatial features for each well in the different regions. A training dataset and a testing dataset are generated by splitting the combined dataset. A machine learning model is trained using cross-validation and tuning on the training dataset to predict estimated ultimate recovery (EUR). The performance of a machine learning (EUR) model is evaluated with respect to different regression metrics. An importance that each of the attributes and features of the well data has on machine learning models is determined.

What is claimed is: 
     
         1 . A computer-implemented method, comprising:
 collecting well data, including attributes and features of basin data, completion data, and production data, for each well in an unconventional field;   generating spatial features for each well in different regions in the unconventional field;   generating, using the well data and the spatial features, a combined well features dataset mapping the well data to the spatial features for each well in the different regions of the unconventional field;   generating, by splitting the combined well features dataset, a training dataset, and a testing dataset;   training a machine learning model using cross-validation and tuning on the training dataset to predict estimated ultimate recovery (EUR);   evaluating the performance of a machine learning (EUR) model with respect to different regression metrics; and   determining, based on the spatial features and by evaluating the EUR model, an importance of each of the attributes and features of the well data in machine learning models, wherein the importance identifies an impact on production by each attribute and feature.   
     
     
         2 . The computer-implemented method of  claim 1 , further comprising:
 cleaning and normalizing the well data to remove well data for wells with missing data and to normalize the well data per requirements for machine learning.   
     
     
         3 . The computer-implemented method of  claim 1 , wherein the spatial features include (x,y) coordinates indicating a horizontal position and a z-coordinate indicating depth. 
     
     
         4 . The computer-implemented method of  claim 1 , wherein the splitting is performed using machine learning modeling and includes applying stratifications based on region having representative well for each regions. 
     
     
         5 . The computer-implemented method of  claim 1 , further comprising using the importance of each of the attributes and features of the well data in designing wells and stimulating wells. 
     
     
         6 . The computer-implemented method of  claim 1 , further comprising determining, using the importance of each of the attributes and features of the well data, horizontal and spatial well placement. 
     
     
         7 . The computer-implemented method of  claim 1 , wherein the spatial features are generated for a time period of 30, 60, or 90 days. 
     
     
         8 . A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:
 collecting well data, including attributes and features of basin data, completion data, and production data, for each well in an unconventional field;   generating spatial features for each well in different regions in the unconventional field;   generating, using the well data and the spatial features, a combined well features dataset mapping the well data to the spatial features for each well in the different regions of the unconventional field;   generating, by splitting the combined well features dataset, a training dataset, and a testing dataset;   training a machine learning model using cross-validation and tuning on the training dataset to predict estimated ultimate recovery (EUR);   evaluating the performance of a machine learning (EUR) model with respect to different regression metrics; and   determining, based on the spatial features and by evaluating the EUR model, an importance of each of the attributes and features of the well data in machine learning models, wherein the importance identifies an impact on production by each attribute and feature.   
     
     
         9 . The non-transitory, computer-readable medium of  claim 8 , the operations further comprising:
 cleaning and normalizing the well data to remove well data for wells with missing data and to normalize the well data per requirements for machine learning.   
     
     
         10 . The non-transitory, computer-readable medium of  claim 8 , wherein the spatial features include (x,y) coordinates indicating a horizontal position and a z-coordinate indicating depth. 
     
     
         11 . The non-transitory, computer-readable medium of  claim 8 , wherein the splitting is performed using machine learning modeling and includes applying stratifications based on region having representative well for each regions. 
     
     
         12 . The non-transitory, computer-readable medium of  claim 8 , the operations further comprising using the importance of each of the attributes and features of the well data in designing wells and stimulating wells. 
     
     
         13 . The non-transitory, computer-readable medium of  claim 8 , the operations further comprising determining, using the importance of each of the attributes and features of the well data, horizontal and spatial well placement. 
     
     
         14 . The non-transitory, computer-readable medium of  claim 8 , wherein the spatial features are generated for a time period of 30, 60, or 90 days. 
     
     
         15 . A computer-implemented system, comprising:
 one or more processors; and   a non-transitory computer-readable storage medium coupled to the one or more processors and storing programming instructions for execution by the one or more processors, the programming instructions instructing the one or more processors to perform operations comprising:
 collecting well data, including attributes and features of basin data, completion data, and production data, for each well in an unconventional field; 
 generating spatial features for each well in different regions in the unconventional field; 
 generating, using the well data and the spatial features, a combined well features dataset mapping the well data to the spatial features for each well in the different regions of the unconventional field; 
 generating, by splitting the combined well features dataset, a training dataset, and a testing dataset; 
 training a machine learning model using cross-validation and tuning on the training dataset to predict estimated ultimate recovery (EUR); 
 evaluating the performance of a machine learning (EUR) model with respect to different regression metrics; and 
 determining, based on the spatial features and by evaluating the EUR model, an importance of each of the attributes and features of the well data in machine learning models, wherein the importance identifies an impact on production by each attribute and feature. 
   
     
     
         16 . The computer-implemented system of  claim 15 , the operations further comprising:
 cleaning and normalizing the well data to remove well data for wells with missing data and to normalize the well data per requirements for machine learning.   
     
     
         17 . The computer-implemented system of  claim 15 , wherein the spatial features include (x,y) coordinates indicating a horizontal position and a z-coordinate indicating depth. 
     
     
         18 . The computer-implemented system of  claim 15 , wherein the splitting is performed using machine learning modeling and includes applying stratifications based on region having representative well for each regions. 
     
     
         19 . The computer-implemented system of  claim 15 , the operations further comprising using the importance of each of the attributes and features of the well data in designing wells and stimulating wells. 
     
     
         20 . The computer-implemented system of  claim 15 , the operations further comprising determining, using the importance of each of the attributes and features of the well data, horizontal and spatial well placement.