Abstract:
This study was aimed at predicting reservoir porosity with the aid of Artificial Neural Network
(ANN) and modeling of reservoir properties vis-à-vis permeability, porosity and water saturation
of 'OSA' field in Niger Delta. In mitigating the challenge caused by inaccurate estimation of
reservoir porosity from well log measurement due to heterogeneity nature of reservoirs, ANN was
employed for its accurate prediction as this technique is independent of inherent uncertainties
present in borehole. The data set used comprises base map, 3-D seismic data, suite of well
logs for five wells, core porosity data of three wells (Wells 1, 2 and 4) and check-shot survey
information of ‘OSA’ field. Well logs data (sonic, resistivity and density) within the reservoir of
interest were selected as input variables of the supervised network while core porosity data of Well
1 was set as its target and trained with MATLAB software based Artificial Neural Network (ANN)
using Levernberg-Marquadt algorithm at epoch 10. The input data were randomly divided into
75% of training, 15% of validation and 15% of testing while the mean squared error and validation
regression obtained after the training were 0.000268816 and 0.999996 respectively. The produced
artificial intelligence script was run to predict porosity (true porosity) of wells 2, 3, 4 and 5.
Detailed petrophysical analysis was also carried out on the well logs using empirical equations to
estimate the reservoir properties for the purpose of modeling. Structural and property modeling
i.e. effective porosity, permeability, water saturation and ANN predicted porosity of reservoir
(RES 1) were distributed stochastically within the constructed 3D framework using Sequential
Gaussian Simulation (SGS). For reservoir (RES 1), effective porosity ranges between 4% and 37%
while ANN predicted porosity is from 14% to 38%, permeability (8900mD - 33000mD) and water
saturation (14%
- 34%). At validation, a very good match was obtained between the ANN predicted porosities and
core porosity data i.e. average deviation error for well 2 and 4 are 0.019 and 0.095 respectively.
This suggests that the ANN predicted porosity model represent the true porosity distribution of
reservoir (RES 1). The result of this study has been able to locate two new hydrocarbon prospect
areas in ‘OSA’ field.
