ESP Choke Control Using LSTM for Pump Intake Pressure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for managing well performance in hydrocarbon production, particularly in multi-zone wells, are time-consuming, subjective, and less accurate, leading to inefficiencies and potential production losses due to unpredictable well performance issues.
Innovation Solution
A machine-learning-based system utilizing a recurrent neural network, specifically a Long Short-Term Memory (LSTM) network, to predict and optimize choke indexer settings for electrical submersible pumps (ESPs) based on static and dynamic well data, enabling real-time adaptability and proactive management of well operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to manage well performance, then the process is simple to implement, but it is time-consuming and less accurate leading to production losses
Solution Approach 1:
The patent replaces traditional mechanical/manual well performance management methods with a machine learning-based system that uses neural networks to automatically analyze well data, predict performance issues, and recommend optimizations. This substitution eliminates time-consuming manual analysis while improving prediction accuracy through computational algorithms.
Solution Approach 2:
The machine learning system performs self-learning and automatic optimization recommendations without requiring continuous manual intervention. The system processes well data autonomously, identifies patterns, predicts performance issues, and generates optimization strategies, thereby reducing the time investment needed from operators while maintaining high accuracy.
2Adaptability or versatility
If traditional well performance management is used, then the system complexity is low, but it cannot provide real-time adaptability and proactive management
Solution Approach 1:
The patent implements a dynamic well performance management system that continuously adapts to changing well conditions in real-time. The machine learning models process incoming well data streams, update predictions, and generate optimized recommendations dynamically, enabling the system to respond to changing reservoir conditions, production rates, and operational parameters without manual reconfiguration.
Solution Approach 2:
The system incorporates feedback mechanisms where well performance data is continuously monitored, compared against predicted values, and used to refine future predictions and recommendations. This closed-loop feedback enables real-time adaptability by learning from actual well behavior and adjusting management strategies accordingly, while the feedback structure is integrated into the existing system architecture.
3Productivity
If manual choke and ESP optimization is performed, then the operational simplicity is maintained, but production losses occur due to unpredictable well performance issues
Solution Approach 1:
The machine learning system performs preliminary analysis of well data to predict potential performance issues before they occur. By identifying trends and patterns that indicate upcoming problems, the system enables proactive adjustment of choke settings and ESP operations to prevent production losses, rather than reacting to issues after they manifest.
Solution Approach 2:
The system optimizes well performance by dynamically adjusting key parameters such as choke index settings and ESP operational parameters based on predicted well behavior. The machine learning models analyze the relationships between these parameters and well performance, generating optimized parameter sets that maximize hydrocarbon recovery while maintaining reliable operation and preventing production losses.
Data Source
AI summary
A method that includes obtaining static and dynamic well data for a well. The static well data describes well design parameters and the static well data describes well properties that change over time. The method includes obtaining first choke index setting data regarding a first choke disposed in a producing zone of the well. The method includes using a recurrent neural network to generate, by a computer processor, first predicted electrical submersible pump (ESP) input data for an ESP in hydraulic connection with the first choke based on the static and dynamic well data, and the first choke index setting data. The method includes determining well performance data for the well based on the predicted ESP input data. The method includes determining well operations for the well based on the well performance data and transmitting a command to a control system that causes the well operations to be performed.