Supervised Self Organizing Maps

The Problem

 In a recent deterministic study carried out for a client it was concluded that the deterministic petrophysical summary simply was not credible. We concluded the evaluation was probably significantly underestimating the volume of sand in the formation system under scrutiny.

The solution

For that reason, it was decided to undertake an analysis based on the Machine Learning (ML) technique referred to as Self Organizing Map (SOM). The outcome is a completely revised understanding of the volume of sand in the reservoir. Moreover, it should significantly improve the information available from the wells 1 and well 2 for making a reservoir model for the prospect.

Example of Self Organizing Maps (SOM) performed with Dtectit app based on the toroidal map geometry. The large map shows the SOM established from mapping of the three lithologies (shale, silt, sand) we want to predict. Maps of the three input variables used are shown also at bottom left. Bottom right shows the training process which is mean distance to closest unit against iteration number.

Scope of work.

  • Facies Analysis from Core
  • Improve the quantification of the amount of sand contained within the Sand 1 and Sand 2 intervals.
  • Extract flow unit information using rock typing and extract beyond the cored intervals to both wells.

*For more information please check the original article here.

Illustration of the predicted Facies (wp_facies) against Facies with Dtectit app using Supervised SOM.
Confusion Matrix shows more than 90% prediction accuracy.

The results of this study show a significant different result compared to the deterministic evaluation. The study concludes that far more sand and fine sand/silt is contained in the formation encountered by the well 1 and well 2 than the deterministic evaluation.