Neuro-fuzzy system to predict permeability and porosity from well log data: A case study of Hassi R'Mel gas field, Algeria

Characterization of shaly sand reservoirs by well log data is a usual way of describing oil/gas field reservoirs. Over the last few years, several studies have been conducted in the field of petroleum engineering by applying artificial intelligence. This work represents a petrophysical-based method that uses well logs and core plug data to predict well log data recorded at depth in a shaly sand reservoir of Triassic Formation in Hassi R'Mel field, Algeria. In the study of oil reservoirs, the prediction of absolute permeability is a fundamental key in reservoir descriptions and has a direct impact, in particular, on effective completion designs, successful water injection programs and more efficient reservoir management. The Triassic Formations of Hassi R'Mel fields are composed of sandstones and shaly sands with dolomites. Logs from 10 wells are the starting point for the reservoir characterization. This paper presents a hybrid neuro-fuzzy model based on the use of data from four wells regarding porosity and permeability estimation. A fuzzy logic approach is used to calibrate the calculated permeability and core permeability; and a neural network was developed in this model, based on the data available from the field. Fuzzy analysis is based on fuzzy logic and is used to choose the best well logs with regard to core porosity and permeability data. A neural network is used as a nonlinear regression method to develop transformation between the selected well logs and core measurements. Porosity and permeability are predicted in these wells through linear regression; and back-propagation models are constructed and their reliabilities are compared according to the regression coefficients for predictions in un-cored sections. This investigative hybrid neuro-fuzzy method becomes a powerful tool for the estimation of reservoir properties from well logs in oil and natural gas development projects. (C) 2014 Elsevier B.V. All rights reserved.