One of the main issues while reservoir geological model creation is to establish the relationship between measured rock properties (e.g. porosity, permeability) and well logs values. Standard approaches consist in using deductive mathematical modeling algorithms to solve this problem. The primary objective of this study was to develop the best mathematical model for Dolgan reservoir rock characteristics estimation using all available well logs information. It is obvious that simple polynomial models for this area were not applicable due to special properties of water saturated. Matched rock properties lab measurements in reference wells and well logs data were available.

In the frameworks of this article several statistical methods including various regression types and neural networks were analyzed. The best relation in terms of statistical criteria was obtained by the group method of data handling (GMDH) method. GMDH is an inductive modeling algorithm using neural network with active neurons. This approach optimizes not only model coefficients for predetermine mathematical equation but also select optimal model complexity.

Multilayered algorithm of GMDH, based on polynomial reference function allowed maximizing amount of information being used from different types of well logs in reference wells for target relation. Data from neutron, density, resistivity and PS logs were the most significant for the final model of Dolgan resorvoir.

So using of GMDH cybernetics algorithms may significantly increase precision of rock properties forecast for further geomodeling purposes.

You can access this article if you purchase or spend a download.