Integrating hydraulic flow unit concept and adaptive neuro-fuzzy inference system to accurately estimate permeability in heterogeneous reservoirs: Case study Sif Fatima oilfield, southern Algeria

Formation characteristics are a crucial requirement for reservoir modeling and they need to be constantly updated as new static and/or dynamic reservoir-property details become available. This improves reservoir interpretation while achieving a good history-matching for a better representation of the reservoir. This paper proposes a new method to estimate reservoir permeability in an oil-producing reservoir by applying two distinct cases: partially cored and completely non-cored wellbores. The approach involves the determination of hydraulic-flow units (HFU) using a flow-zone indicator (FZI) distribution to consider the dynamic aspect of the permeability, which strongly influences fluid flow through porous media. The FZI is clustered by normal probability analysis to identify the optimum number of HFU clusters. The estimated reservoir permeability is validated in static and dynamic reservoir models. For an accurate prediction of the FZI over the entire reservoir thickness, an adaptive-neuro-fuzzy-inference system (ANFIS) is developed using conventional well-log variables: (i) gamma-ray (GR), (ii) sonic (DT), (iii) density (RHOB), (iv) deep resistivity (DLL), and (iv) neutron porosity (NPHI) calibrated with the available core data. The static regression parameters root-mean-squared error (RMSE), mean absolute error (MAE), coefficient of determination (R2), and R2adjusted are used to evaluate the performance of the ANFIS model in predicting FZI to determine HFU for estimating reservoir permeability. To validate this approach, a matching quality check of the downhole and reservoir pressure is carried out using a dynamic reservoir simulation model. Historical data collected from wellbore OPW-1 drilled in the Sif Fatima oilfield (Algeria) are used to evaluate the accuracy of the proposed permeability-prediction framework. The ANFIS model can accurately and efficiently predict the FZI exhibiting R2, R2adjusted, RMSE, and MAE of approximately (≈0.981, ≈0.979, ≈0.014, ≈4,600E-08)), (≈0.933, ≈0.891, ≈0.032, ≈1.615E-03), and (≈0.970, ≈0.968, ≈0.018, ≈2.381E-04)) for the training, testing and complete dataset, respectively. When applied to predict reservoir permeability, the Hydraulic Flow Unit concept was able to generate accurate results from the predicted FZI data by applying the modified Kozeny-Carman equation with, R2, RMSE and MAE values of approximately ≈0.9702, ≈0.969, ≈9.604, and ≈4.398, respectively. The calculated permeability was validated successfully at well and field levels. Additionally, it generated more accurate history matches of the well and the reservoir pressure trends than the previous permeability model, and consequently improved the reservoir modelling workflow.