Approaches to needle valve optimization

Abstract

This disclosure relates to methods, systems, and computer-readable media for collecting and analyzing historical data related to pressure buildup and gas production for use in training and implementing one or more models for predicting the relationship between gas production and needle valves. In particular, this disclosure relates to training an implementing a pressure buildup and production model to be used in connection with determining when to open a needle valve that controls the flow of gas within a wellsite environment. The following disclosure describes features and characteristics related to specifically training a number of machine learning models and, based on a combination of predictions and real-time production and pressure data, further training the machine learning models to recommend a timing when the needle valve should be open to maximize production of the wellsite over time.

What is claimed is: 
     
         1 . A method, comprising:
 training a pressure buildup and production model to generate a given output indicating a predicted production metric for an upcoming production period upon opening a needle valve to initiate gas production of a well, wherein the pressure buildup and production model is trained based on historical production data and historical pressure data;   obtaining input pressure and production data for previous production periods and previous closing periods;   applying the pressure buildup and production model to the input pressure and production data to generate an output indicating a predicted production metric for a next production period of a wellsite; and   causing the needle valve to open based on the predicted production metric for the next production period indicated by the output of the pressure buildup and production model.   
     
     
         2 . The method of  claim 1 , wherein the pressure buildup and production model includes:
 a pressure buildup model comprising a deep neural network, the pressure buildup model being trained to generate a first output indicating a predicted pressure buildup metric based on a first input comprising a first pressure metric associated with a previous valve closing period; and   a production model comprising a tree-base machine learning model, the production model being trained to generate a second output comprising a predicted production metric for a next production period based at least in part on the first pressure metric output by the pressure buildup model.   
     
     
         3 . The method of  claim 2 , wherein the production model generates the output of the pressure buildup and production model based at least in part on a predicted pressure metric generated by the pressure buildup model. 
     
     
         4 . The method of  claim 2 , wherein the pressure buildup metric is a measurement of pressure at a time corresponding to when the needle valve is closed at a beginning of the previous valve closing period. 
     
     
         5 . The method of  claim 2 , wherein the pressure buildup metric is measurements of pressure over a range of time corresponding to the valve closing period. 
     
     
         6 . The method of  claim 2 , further comprising:
 measuring gas production of the next production period;   measuring pressure buildup during the valve closing period;   adding the measured gas production and the measured pressure buildup to a database including the historical production data and historical pressure data; and   retraining the pressure buildup and production model based on the measured gas production added to the database.   
     
     
         7 . The method of  claim 6 , wherein retraining the pressure buildup and production model includes:
 retraining the deep neural network based on the measured pressure buildup; and   retraining the tree-based machine learning model based on a combination of the measured pressure buildup and the measured gas production.   
     
     
         8 . The method of  claim 1 , wherein the historical production data and historical pressure data comprises:
 measured values of tubing holding pressure (THP) and casing holding pressure (CHP) over two or more previous production cycles including valve closing periods and associated production periods; and   measure gas production over the two or more previous production cycles.   
     
     
         9 . The method of  claim 1 , wherein the output of the pressure buildup and production model indicates a recommended time for opening the needle valve at an end of the closing period to maximize gas production over multiple production periods including the next production period. 
     
     
         10 . A system, comprising:
 at least one processor;   memory in electronic communication with the at least one processor;   instructions stored in the memory, the instructions being executable by the at least one processor to:
 train a pressure buildup and production model to generate a given output indicating a predicted production metric for an upcoming production period upon opening a needle valve to initiate gas production of a well, wherein the pressure buildup and production model is trained based on historical production data and historical pressure data; 
 obtain input pressure and production data for previous production periods and previous closing periods; 
 apply the pressure buildup and production model to the input pressure and production data to generate an output indicating a predicted production metric for a next production period of a wellsite; and 
 cause the needle valve to open based on the predicted production metric for the next production period indicated by the output of the pressure buildup and production model. 
   
     
     
         11 . The system of  claim 10 , wherein the pressure buildup and production model includes:
 a pressure buildup model comprising a deep neural network, the pressure buildup model being trained to generate a first output indicating a predicted pressure buildup metric based on a first input comprising a first pressure metric associated with a previous valve closing period; and   a production model comprising a tree-base machine learning model, the production model being trained to generate a second output comprising a predicted production metric for a next production period based at least in part on the first pressure metric output by the pressure buildup model.   
     
     
         12 . The system of  claim 11 , wherein the production model generates the output of the pressure buildup and production model based at least in part on a predicted pressure metric generated by the pressure buildup model. 
     
     
         13 . The system of  claim 11 , wherein the pressure buildup metric is a measurement of pressure at a time corresponding to when the needle valve is closed at a beginning of the previous valve closing period. 
     
     
         14 . The system of  claim 11 , wherein the pressure buildup metric is measurements of pressure over a range of time corresponding to the valve closing period. 
     
     
         15 . The system of  claim 11 , wherein the instructions are further executable by the at least one processor to:
 measure gas production of the next production period;   measure pressure buildup during the valve closing period;   add the measured gas production and the measured pressure buildup to a database including the historical production data and historical pressure data; and   retrain the pressure buildup and production model based on the measured gas production added to the database.   
     
     
         16 . The system of  claim 15 , wherein retraining the pressure buildup and production model includes:
 retraining the deep neural network based on the measured pressure buildup; and   retraining the tree-based machine learning model based on a combination of the measured pressure buildup and the measured gas production.   
     
     
         17 . The system of  claim 10 , wherein the historical production data and historical pressure data comprises:
 measured values of tubing holding pressure (THP) and casing holding pressure (CHP) over two or more previous production cycles including valve closing periods and associated production periods; and   measure gas production over the two or more previous production cycles.   
     
     
         18 . The system of  claim 10 , wherein the output of the pressure buildup and production model indicates a recommended time for opening the needle valve at an end of the closing period to maximize gas production over multiple production periods including the next production period. 
     
     
         19 . A non-transitory computer readable medium storing instructions that, when executed by at least one processor, causes one or more computing devices to:
 train a pressure buildup and production model to generate a given output indicating a predicted production metric for an upcoming production period upon opening a needle valve to initiate gas production of a well, wherein the pressure buildup and production model is trained based on historical production data and historical pressure data;   obtain input pressure and production data for previous production periods and previous closing periods;   apply the pressure buildup and production model to the input pressure and production data to generate an output indicating a predicted production metric for a next production period of a wellsite; and   cause the needle valve to open based on the predicted production metric for the next production period indicated by the output of the pressure buildup and production model.   
     
     
         20 . The non-transitory computer readable medium of  claim 19 , wherein the pressure buildup and production model includes:
 a pressure buildup model comprising a deep neural network, the pressure buildup model being trained to generate a first output indicating a predicted pressure buildup metric based on a first input comprising a first pressure metric associated with a previous valve closing period; and   a production model comprising a tree-base machine learning model, the production model being trained to generate a second output comprising a predicted production metric for a next production period based at least in part on the first pressure metric output by the pressure buildup model.