Machine Learning Well Production Forecasting

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for predicting hydrocarbon well production in unconventional formations face challenges due to limited data availability and the inefficiencies of traditional decline curve analysis, which can lead to inaccurate forecasts and increased uncertainty in production estimates.

Innovation Solution

A computer-implemented method utilizing machine learning techniques, including data analytics and regression models, to forecast hydrocarbon well production by training models on historical data and using ensemble techniques to generate production predictions for future intervals, even with limited initial production data, thereby improving prediction accuracy and reducing computational time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional decline curve analysis is used for production forecasting, then the method is simple and easy to implement, but the prediction accuracy is low and uncertainty is high

Engineering Contradiction:
Improveprediction accuracyVSAvoidmethod complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transforms the forecasting approach by changing from traditional decline curve parameters to machine learning model parameters. It uses multiple input parameters including well completion data, formation properties, and production history to train regression models, thereby improving prediction accuracy while managing complexity through automated computational processes.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces the mechanical decline curve analysis method with a data-driven machine learning system. It uses ensemble regression models that process multiple factors and variables simultaneously, substituting the simple but inaccurate mechanical approach with a more complex but accurate computational intelligence system.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If more historical data is collected for training models, then the prediction accuracy improves, but the data processing time and computational resources increase

Engineering Contradiction:
Improveforecast accuracyVSAvoidcomputational time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-training the machine learning models on extensive historical data before actual forecasting is needed. The ensemble regression models are trained offline on available production data, completing the heavy computational work in advance so that subsequent predictions can be generated rapidly without requiring extensive processing time.

Inventive Principle:
Principle #10Preliminary action

3Reliability

If traditional methods are used with limited initial production data, then the process is straightforward, but the forecast reliability is poor

Engineering Contradiction:
Improveforecast reliabilityVSAvoidmodel complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces machine learning regression models as intermediaries between limited initial production data and reliable forecasts. These models act as mediators that incorporate additional information from well completion data, formation properties, and analogous well performance to compensate for the limited production history, thereby improving forecast reliability without requiring extensive direct production data.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3406844B1Resource production forcasting
Publication Date: 2024.06.12 SENSIA NETHERLANDS BV
  • EP3406844B1 patent drawingFigure 1
  • EP3406844B1 patent drawingFigure 2
  • EP3406844B1 patent drawingFigure 3

AI summary

A method includes receiving data where the data include data for a plurality of factors associated with a plurality of wells; training a regression model based at least in part on the data and the plurality of factors; outputting a trained regression model; and predicting production of a well via the trained regression model.