Machine Learning Assisted Well Completion Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional history matching techniques for wellbore models are time-consuming and unable to effectively incorporate all available production history data, leading to inefficiencies in well planning and production forecasting, and the knowledge gained from one wellbore is not transferable to new wellbores, requiring repetitive history matching processes.
Innovation Solution
The implementation of machine learning (ML) assisted history matching using artificial intelligence and deep neural networks to correlate geological characteristics and historical production data from existing well sites to estimate history-matched modeling parameters for new well sites, enabling more accurate production forecasting before drilling commences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional history matching techniques are used to tune wellbore model parameters, then model accuracy is improved, but the time required for the process increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models using historical well data before actual history matching is needed. The ML models are trained offline on extensive datasets of wellbore parameters and production data, so that when history matching is required for a new well, the parameter estimation can be performed rapidly using the pre-trained model rather than starting from scratch with conventional iterative methods
Solution Approach 2:
The patent substitutes the mechanical iterative optimization process of conventional history matching with a machine learning-based parameter estimation system. Instead of repeatedly running full reservoir simulations and adjusting parameters through trial and error, the system uses trained ML models (including neural networks) to directly estimate optimal parameters from well data, dramatically reducing computational time while maintaining accuracy
2Measurement precision
If conventional history matching techniques are applied to each new wellbore, then accurate production forecasts are obtained, but the process must be repeated for every well, increasing overall workload
Solution Approach 1:
The patent implements universality by creating machine learning models that can be applied across multiple wellbores. The ML models are trained on data from multiple wells and can then be used to estimate parameters for new wells in the same reservoir or similar geological settings. This universal approach allows the system to maintain accurate production forecasts for each well while avoiding the need to repeat the entire history matching process from scratch for every new wellbore
Solution Approach 2:
The patent applies copying by using knowledge and parameter estimates derived from historical well data to inform the modeling of new wells. The ML models capture patterns and relationships from existing well data that can be transferred and applied to predict parameters for new wells, effectively copying useful information across different wellbores rather than treating each well as entirely independent
3Reliability
If comprehensive historical production data is incorporated into the model, then model reliability is improved, but the complexity of the history matching process increases
Solution Approach 1:
The patent applies the extraction principle by using machine learning models to automatically identify and extract the most relevant features and patterns from comprehensive historical production data. Rather than requiring manual processing of all available data, the ML models automatically extract key information and relationships, reducing the effective complexity of working with large datasets while maintaining model reliability through the use of all available historical information
Data Source
AI summary
Systems and methods for completion design are disclosed. Wellsite data is acquired for one or more existing production wells. The wellsite data is transformed into model data sets for training a first machine learning (ML) model to predict well logs. A first well model uses the well logs to estimate production of the existing well(s). Parameters of the first well model are tuned based on a comparison between the estimated and actual production of the existing well(s). A second ML model is trained to predict parameters of a second well model for a new well, based on the tuned parameters of the first well model. The new well's production is forecasted using the second ML model. Completion costs for the new well are estimated based on the well's completion design parameters and the forecasted production. Completion design parameters are adjusted, based on the estimated completion costs and the forecasted production.