Hydraulic unit prediction using support vector machine

Hydraulic Flow Units or hydraulic units (HUs) concept is becoming popular for grouping reservoir rocks of similar petrophysical properties with the main goal of having a better estimate of permeability. HU approach has an advantage that it addresses the development of permeability in reservoir rocks from fundamentals of geology and physics of flow at pore network scale. The aim of the present study is to predict HUs for the un-cored sections of the wells in a carbonate reservoir using Support Vector Machines (SVMs). HUs for un-cored sections were predicted using wire line logs as input and the associated conventional core values as guides to SVMs. HUs for the core data were identified using three popular correlations such as Kozeny–Carmen (KC) equation, Nooruddin and Hossain equation, and the power law flow unit equation. The experimental results on a Middle East field data show that a better HU prediction accuracy is achieved using Nooruddin and Hossain correlation in comparison with using KC and the power law flow unit correlation. A further analysis to the predicted value reveals that better prediction accuracy is achieved if the granularity of HU class boundary is enlarged to the neighboring classes. Although Nooruddin and Hossain correlation could better relate wire line logs to HUs, however the permeability calculated from the predicted HUs showed less error with the power law flow unit correlation. Considering this, we achieved a maximum of 97% accuracy which encourages the application of SVMs in HU unit prediction using well log data for the un-cored sections of well or for the wells which do not have core data.