Machine Learning Production Prediction for Coal-Bed Methane Wells
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Estimating the volumes of gas and water production from coal-bed methane (CBM) wells is challenging due to the complexities in coal-bed methane production, including large internal surface areas and the need for accurate planning and well construction to meet delivery demands, which existing methods fail to address effectively.
Innovation Solution
A computer-implemented method that combines field data and software tools for analyzing production data using machine learning to predict gas and water production quantities directly for new wells, bypassing the clustering phase, and providing a production prediction curve based on well parameters and historical data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physically-based probabilistic well models are used for production estimation, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent replaces complex physically-based probabilistic models with a machine learning system that uses historical production data and well parameters to predict future production. The machine learning model learns patterns from data without requiring explicit physical equations, thereby reducing model complexity while maintaining or improving prediction accuracy.
Solution Approach 2:
The patent uses historical production data from existing wells to create a data-driven model that copies successful production patterns. By training on historical data, the system learns to predict production for new wells without requiring complex physical modeling of each well scenario.
2Measurement precision
If clustering phase is included in production prediction methodology, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent extracts and removes the clustering phase from the traditional production prediction workflow. By eliminating this intermediate step, the system achieves faster predictions while maintaining accuracy through direct machine learning modeling that processes well parameters and historical data without requiring separate clustering operations.
Solution Approach 2:
The patent skips the clustering phase entirely by using a machine learning model that directly maps well parameters to production predictions. This allows the system to rush through the prediction process more quickly by bypassing unnecessary intermediate computational steps.
3Productivity
If more wells are constructed to meet delivery demands, then productivity is improved, but loss of substance increases
Solution Approach 1:
The patent uses machine learning models trained on historical production data to predict both gas and water production for each well scenario. This feedback mechanism allows operators to evaluate the water-gas ratio for different well construction plans and select configurations that maximize gas delivery while minimizing water production, thereby resolving the contradiction between productivity and substance loss.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Estimating a production prediction of a target well includes computing, based on production time series from training wells, a smoothed production history curves. Each smoothed production history curve corresponds to a training well. Based on the smoothed production history curves, a fitting function defined by a set of fitting coefficients is selected. A machine learning process determines, based on a set of well parameters for each training well, a set of predicted fitting coefficients as a function of a set of well parameters of the target well. Estimating the production prediction further includes applying the predicted fitting coefficients to the fitting function to compute a production prediction curve for the target well, and presenting the production prediction curve.