Machine learning-assisted history matching
Challenge
- Reducing time and costs during the history matching study phase.
- Building models with better predictive power and ability for comprehensive quantification of uncertainty.
- Modeling complex reservoirs with long production histories undergoing different recovery mechanisms and sharing the same surface facility.
- Traditional history matching process is very manual, tedious and takes months to complete.
- History matching workflow is not integrated with geomodeling and field development planning (FDP) processes.
- Lack of time and computing resources leads to a single simulation case forecast which does not address all possibilities.
Solution
Our ML-assisted history matching technique has been developed to find the optimum values of subsurface parameters. It uses the proxy modeling approach, combined with ML to solve the complex relationships between target variables, objective function and multiple independent variables, and reservoir uncertainties.
“Previously, our workflow took three to six months—it can now be shortened to a one-month study. It’s now automatic and robust enough to deliver our next development plan.”
Digital subsurface manager
Results
Fully integrated propagation of static and dynamic uncertainties.
Probabilistic approach to uncertainty quantification.
Customizable ML-proxy for affordable generation of the conditioned posteriors.
Fully integrated decision system for confident decisions.
Improved model quality and prediction accuracy.
Improved productivity reducing time to decisions.
