Abstract:
A Probabilistic Neural Network (PNN) multiattribute transform was applied to a three –
dimensional (3D) seismic data and a suite of borehole log data obtained within ‘Bimbo’ field,
offshore Niger Delta. This was done to characterize the delineated reservoirs in terms of the
predicted porosity, density and acoustic impedance variations across the reservoirs. Tops of
hydrocarbon-bearing reservoirs were identified on the well logs which were tied to the seismic
data with the aid of a synthetic seismogram A post stack seismic inversion was carried out to create
an impedance volume which served as an external attribute for the Multi-linear attribute regression
analysis. The multi-attribute linear regression analysis determined the optimum group of seismic
attributes to be used for the Probabilistic Neural Network which was trained to generate a nonlinear
relationship between the specified well log petrophysical parameters (density and porosity)
and the seismic attributes. The established relationship was used to predict the petrophysical
parameters of interest across the study area through which hydrocarbon prospect zones were
identified. Three (3) hydrocarbon-bearing sands, - Res_A, Res_B and Res_C, - were delineated
from the well log data and the petrophysical parameters were calculated. Application of the
Probabilistic Neural Network in predicting porosity provided a correlation coefficient of 85%
while its validation result provided a correlation coefficient of 75%. For Density prediction, the
Probabilistic Neural Network provided a correlation coefficient of 90% while its validation result
provided a correlation coefficient of 87%. Integration of the PNN-predicted petrophysical maps
(Porosity, Density and Acoustic impedance) for the tops of Res_B and Res_C revealed two
prospect areas ‘X’ and ‘Y’ characterized by low acoustic impedance values, high porosity and low
density. The study has established a reliable relationship between log-derived petrophysical
parameters of the reservoirs and seismic attributes.
