• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Research in oil field development forecasting for Schlumberger

 

Oil production forecasting is a problem of major importance in petroleum. Given oil field structure and geology, we want to receive our performance for any chosen scenario of development even before we drill the wells.

Nowadays, one of the most popular strategies for oil-producing is to drill a number of production wells and injection wells which allow us to reservoir pressure supporting. There may be a lot of scenarios with different water injection schedules, production wells equipment, well completion methods etc. Moreover, different oil reservoirs will behave differently in equal conditions.

Nowadays, commonly used method is to divide our reservoir into a grid and evaluate flows between conjugated blocks for each fluid present in the rock using finite differences technique. Such simulators assume that rock properties inside an arbitrary block are constant. They also compute pressure and saturation changes and a lot of other stuff.

Although these simulators are quite accurate, they have two main cons:

- One scenario of around three to five years length may take weeks of evaluation in case of a huge oil field.
- One must consider properties for every each block of the grid which is not possible due to the complexity of property exploration in such depths. One can use interpolation techniques, but this method can not use information from explored production data.

Bayesian group’s members under the leadership of Dmitry Vetrov and SkolTech University’s students under the leadership of Ivan Oseledets are working on the project that addresses these two cons.

Our aim is to construct a new system, based on complex deep neural network and, possibly, some bayesian inference techniques, that will work faster than finite differences simulators, will need no parameters chosen by hand and will have at least the same accuracy as a finite differences ones. Our model will be rather data-driven, than physical, but it will involve most major physical properties of the oil field system.

The input data will consist of only injection schedule, production well’s equipment information and information about oil production intensification operations for any concrete oil field.

In comparison with other existing data-driven techniques, our method will provide the solution for more development scenario types, in different oil field’s life stages and even in the case of intensification operations such as hydraulic fracturing and acid treatment. This was never done before.


 

Have you spotted a typo?
Highlight it, click Ctrl+Enter and send us a message. Thank you for your help!
To be used only for spelling or punctuation mistakes.