Characterization of the reservoir is very time-consuming task, which requires expertise of many engineers.
No Direct Measurements
Reservoir properties are measured only at the well locations, while reservoir model requires continuous spatial distribution of the properties for description of the reservoir behavior.
For many petrophysical properties there are no direct measurements, therefore the necessary correlations should be derived.
Supervised ML for Spacial PropertiesSupervised machine learning techniques for the prediction of spatial properties (such as permeability, porosity, fracture intensity, etc.) are used in combination with Self-Organizing Maps (SOMs), which are applied to delineate lithofacies from well log data. Further on either cluster membership from the SOM model is used as additional input for the predictive models, or predictive models are developed for each cluster separately.
Added Value Gained
Improved ModelImproved reservoir characterization model
- Canadian CHOPS operator
- Major European operator