Oil reservoir classification for ultimate oil recovery estimation by means of machine learning

Oil recovery estimation is the most important tasks after calculation of oil in place and thereafter in oil development plans. There are a lot of appropriate methods for such estimation - displacement coefficient, sweep efficiency, waterflood efficiency, using final well water cut, with respect to fluid mobilities, reservoir thickness and porosity, absolute and relative permeability. Often such parameters are taken from similar nearest reservoirs due to lack of the data. Reservoir simulation is another method for oil recovery estimation although it has many shortcomings. Oil recovery estimation presented in this paper is based on widely known k-means unsupervised machine learning algorytms. Silhouette technics is used for choosing main clusters. Parameter euristics based on local Volga-Ural region data is diveded by clusters for oil recovery. Reservoir classification methodology can dramatically improve ultimate recovery estimation.