The application of data clustering on the basis of Kohonen self-organizing maps in the process of selecting candidate wells for enhanced oil recovery methods

A methodology of preliminary clustering of field information is suggested on the basis of Kohonen self-organizing maps within the process of selecting candidate wells for enhanced oil recovery methods. The peculiarity of modeling problems of estimation of the technological efficiency of improving water injection profile is the high degree of differentiation of initial information, which is caused by the compartmentalization of formation characteristics (porosity, permeability, lateral continuity etc.), the different well characteristics (rates, bottom hole pressure values), the degree of efficiency of formation pressure maintenance in the production zone etc. Therefore, by creating an artificial neural network model based on ungrouped initial information, the average error of oil-production well rate prediction after improving water injection profile was over 50 per cent. To improve the prediction ability of a neural network, the initial data for training the network have been revised and preliminary grouping initial information by means of the Kohonen self-organizing maps has been done. It has been shown that preliminary grouping improved the prediction efficiency of a neural network nearly twice. The average error in the well water-cut prediction is 6.5 per cent. The suggested method is characterized by the far lesser work-time in comparison with the hydrodynamic modeling.

Keywords

Authors

References