Research Projects and Independent Studies
This study evaluates the effectiveness of artificial neural networks and sequential Gaussian simulation in estimating permeability from well log data and predicting values in unsampled reservoir sections. Well log inputs, such as spontaneous potential and resistivity readings, were used as input for the neural network model. At the same time, variography analysed spatial characteristics for the SGS method, with a vertical semi-variogram algorithm applied to remove sedimentation-induced trends. The neural network trained using the Levenberg-Marquardt optimisation algorithm, outperformed SGS by achieving a lower root mean square error of 0.0449 compared to 0.1789 for SGS, due to its ability to capture non-linear relationships between inputs and outputs. Despite SGS's limitations in heterogeneous regions, it maintained spatial consistency by relying on spatial correlations. We suggest that combining both methods could offer a more robust approach, leveraging the neural network accuracy in capturing complex non-linear interactions and SGS's spatial consistency, thereby enhancing reservoir characterisation in areas with limited direct sampling and improving the prediction of reservoir behaviours.
Workflow used for designing neural network model
This study reconstructs the depositional environment of hydrocarbon-bearing formations by analysing gamma-ray logs and 3D seismic data from six wells, identifying sequence patterns—progradational, retrogradational, and aggradational—that reflect energy conditions typical of a deltaic environment.
Gamma-ray logs were used to identify lithologies such as sandstones, shales, and transitional types. Log motifs like bell, funnel, and cylindrical shapes indicate various depositional processes and environments such as fluvial channels and deltaic distributaries. Cross-sectional models correlated lithologies between wells, revealing a consistent and connected reservoir system with significant hydrocarbon potential in the basinward direction. Seismic reflection data further supported these findings by revealing aggradational reflectors and lateral continuity of lithological units, with stacking patterns linked to depositional energy conditions. The prevalence of cylindrical log motifs suggested a uniform and aggrading depositional sequence across the wells. This integrated approach effectively reconstructed the depositional environment, highlighting both the hydrocarbon potential and the energy conditions during deposition.