Machine learning history matching
Challenge
- Reservoir engineers often find most history matching tools difficult to use.
- History matching can be a very time-consuming process.
- Large models and well count increase simulation runtimes and makes history matching very expensive.
Solution
This is a Petrel plug-in designed to expedite the history matching process by leveraging AI to create precise ML-based reservoir proxy models. These models effectively capture the non-linearities of simulation responses, enabling quicker calibration of large, complex reservoir models with numerous wells and extensive production histories.
- Manages Uncertainty: Incorporates uncertainty study results from reservoir simulations to train, validate, and test the ML model. This model acts as a proxy, predicting production, injection, and pressure profiles for all wells based on history matching parameters.
- Optimized Realizations: Optimizes using these ML proxies to minimize mismatches and generate one or multiple history-matched realizations.
Results
Risk quantification based on the probabilistic evaluation of field development options.
History matching speed increased more than 50-fold.
Automated validation of optimized cases.
