An adaptive neuro-fuzzy inference system for prediction of hydraulic flow units in uncored wells: a carbonate reservoir

Abstract

ABSTRACTGeoscientists have always sought various approaches to improve reservoir characterisation by compartmentalising the depth interval into subsections with the highest consistency in pore throat size and distribution. Hydraulic flow units have demonstrated success in segmenting the depth interval of interest into subsections with distinguishable rock and fluid properties. At the primary stage, flow zone indicator values are calculated from core data within the reservoir of interest. A flow zone indicator is an acceptably unique measurement of the flow character of a reservoir interval, giving the relationship between petrophysical properties at the pore scale, like tortuosity and surface area, and the formation scale, say porosity and permeability. The next step segments the reservoir into more accurately delineated depth intervals, or hydraulic flow units. Several statistical approaches have been used successfully to group data into subsections of high similarity and consistency, herein referred to as hydraulic flow units. In this paper, a robust method was proposed for the prediction of flow zone indicators in uncored wells, which may lead to advances in reservoir management, saving a considerable amount of revenue merely by accurately predicting depth interval flow properties without the need for expensive coring operations. The results of this study show that adaptive neuro-fuzzy inference systems can be used with higher levels of confidence to model the unknown, but invaluable, data in uncored but logged wells. The results of this study proved the success of machine-learning approaches in identifying underlying trends and relationships within the data, as well as predicting unknown properties based on training data validated by blind test data. This study shows that soft computing and machine-learning approaches can be used to prognosticate the underlying hydraulic flow units based on well log responses in carbonate reservoirs.