History matching through dynamic decision-making.

Area of Science:

• Reservoir Engineering
• Machine Learning
• Optimization

Background:

• History matching is crucial for reservoir management, using uncertain model attributes to replicate real reservoir performance.
• Accurate reservoir models inform critical decisions in economic analysis and production strategies.
• Traditional methods can be computationally intensive, requiring numerous simulations.

Purpose of the Study:

• To introduce a dynamic decision-making optimization framework for reservoir history matching.
• To leverage machine learning for a data-driven approach to generate improved history-matched models.
• To enhance the efficiency and accuracy of the history matching process.

Main Methods:

• Developed a dynamic optimization framework incorporating a 'learning-from-data' strategy.
• Integrated two machine learning-based optimizer components: unsupervised learning and statistical analysis.
• Utilized pattern recognition from available solution data to guide the generation of new models.
• Applied the framework to the UNISIM-I-H benchmark model.

Main Results:

• The dynamic decision-making optimization framework demonstrated improved quality of history matching solutions.
• The proposed method achieved comparable or better results with a significantly reduced number of simulations compared to previous work.
• Machine learning components effectively identified input attribute patterns linked to favorable reservoir responses.

Conclusions:

• The dynamic decision-making optimization framework offers a more efficient and effective approach to reservoir history matching.
• This data-driven, machine learning-enhanced methodology holds significant potential for advancing reservoir engineering practices.
• The framework's ability to reduce simulation requirements presents a substantial advantage for practical reservoir management.