Prediction of Wells Productive Characteristics with the Use of Unsupervised Machine Learning Algorithms

Abstract

The new approach for production prediction was developed and is described in the article which involves the clustering analysis aimed to well logs such that the reservoir and non-reservoir rocks are obtained (presented by various clusters) and the subsequent linkage among clusters' types, their thicknesses and production characteristics is found. It may be implemented for the prediction of the production for planned wells' ranking further. Such approach may provide the solution to various tasks. It may be used for the geological features' estimation. For example, clustering may be adjusted such that sedimentological interpretation becomes simpler, geological models may become more accurate since the machine interpretation was confirmed to be able to find the errors in human interpretation. Moreover, the necessity of implementation of the method was proved for production prediction by regression analysis when clustering results and dynamic characteristics are used simultaneously.