Neural Network and Empirical Models of Mamuniyat Reservoir Permeability Prediction, Murzuq Basin-Libya

Permeability (K) is a dominant property of the production and development planning strategy of oil fields. Thus, deriving and/or defining a suitable reservoir permeability model provides time and cost consumption. Most reservoir characterization studies were associated with the estimation of permeability. In the present work, three approved empirical model cases were employed by Wyllie and Rose (1950), Timur (1968), and Sheffield (1956). These models' worth utilization to predict the permeability property of the Mamuniyat reservoir (Upper Ordovician) in Murzuq Basin, SW Libya. In addition, derivation of the permeability model of the studied reservoir is based on routine core analysis (CCA) data of two oil wells. Also, a neural network (NNW) is applied to assess the prediction permeability, dependent on measurements of well logging data. Whereas validation of the predicted permeability model is taken into consideration when adding two more oil wells, that are producing from the same reservoir. In general, the predicated permeability values of the clastic Mamuniyat reservoir are greater than 0.1 mD and do not exceed 200 mD, with a good effective porosity (Øe ≈ 13%). However, correlation between the predicted permeability results by the empirical and derive models are satisfactory, with a significant level (P) equal to 0.000. Furthermore, a statistical analysis emphasis both the Derived and NNW models, which hold a regression coefficient (R2) close to 1. Moreover, the grain size and irreducible water (Swi) have an impact on the predicted permeability.