Analysis of permeability based on petrophysical logs: comparison between heuristic numerical and analytical methods

Abstract A series of novel heuristic numerical tools were adopted to tackle the setback permeability estimation in carbonate reservoirs compared classical methods. To that end, a comprehensive data set petrophysical including core and log two wells was situated Marun Oil Field. Both wells, Well#1 Well#2 , completed Bangestan reservoir, having broad diversity facies. In light high Lorenz coefficients, 0.762 0.75 respectively, an extensive heterogeneity has been expected reservoir properties, namely permeability. Despite used as blinded well, which had no influence on model learning just contributed assess validation proposed model. An HFU with aim discerning sophistication net porosity interrelation developed framework Amaefule’s technique modified by newly introduced classification clustering conceptions. Eventually, seven distinct pore geometrical units have distinguished through implementing hybridized genetic algorithm k -means algorithm. Furthermore, K-nearest neighbors (KNN) carried out divide into flow assigns them pre-identified FZI values. Besides, cross between ε-SVR model, supervised machine, Harmony Search estimate directly select optimum combination involved logging parameters reduce dimensionality problem, principle component analysis (PCA) implemented set. The result PCA illustrates parameters, such permeability, transit time sonic wave, resistivity unflashed zone, neutron porosity, photoelectric index, spectral gamma-ray, bulk density, possess highest correlation coefficient first derived PC. line previous studies, findings will be empirical methods, Coates–Dumanior, Timur both launched these wells. Overall, it is obvious conclude ε - SVR undeniably superior method lowest mean square error, nearly 4.91, R-squared approximately 0.721. On contrary, transform relationship remarkably worst results comparison other models error (MSE) accuracy ( R 2 ) 128.73 0.116, respectively.