Integrated application of seismic inversion and artificial neural networks for reservoir characterization and prospect de-risking of OA-field.
The study modified seismic inversion algorithm which incorporated artificial neural networks for reservoir characterization to pre-stack seismic data acquired in the “OA”-Field. The adopted workflow was used to transform the seismic data to acoustic impedance and petrophysical attribute “SoPhi”; considered to be a direct hydrocarbon indicator, in an integrated inversion scheme. This was with a view to characterizing “OA”-Field reservoirs, identifying the exploratory prospects and proposing possible development opportunities. The data used for the study included pre-stack seismic substacks (near-, mid-, and far) and data for one well “OA” comprising gamma ray, resistivity, density, sonic, caliper, and neutron logs. Schlumberger’s Petrel and CGG Veritas’s Hampson Russell softwares were used for the analyses of the data. The hydrocarbon-bearing reservoirs were identified from the well logs using petrophysical parameters such as porosity, water saturation and the volume of shale. Well-to-Seismic calibration was used to tie the formation tops to seismic, thereafter, faults and horizons interpretation were carried out on the pre-stack seismic data. Wavelets were extracted from all the seismic substacks, while a priori impedance (Zp, Zs) and density models were generated from the P-impedance, computer S-impedance (using a modified form of Castagna’s equation) and density logs, within a time window that covered the full extent of all observed hydrocarbon bearing intervals. The pre-stack seismic inversion workflow was run on the initial model for the selected time window and was progressively iterated over 100 runs to minimize the residual difference between the inverted model and the original seismic data. The inversion results with other seismic attributes were then fed as input into a back propagation neural network (BPNN) algorithm which learned the relationship between the desired output and the training data. This was then used to convert the original seismic data into a hydrocarbon presence indicator. Four hydrocarbon bearing reservoirs (A, B, C and D) were delineated from the OA well log. Reservoirs A and B were oil bearing, C was gas and oil bearing while reservoir D was gas-filled. Average porosity progressively decreased from Reservoir A (23%) to the deepest Reservoir D (at 16%). Hydrocarbon saturations of the reservoirs A, B, C and D were 55%, 66%, 58% and 68% respectively. Six faults (A, B, C, D, E and F) and seven horizons (HOR A, B, C, D, E, F and G) were interpreted in the “OA”-Field. Horizons E and F defined the upper and lower boundaries of the interval of the seismic inversion. The pre-stack seismic inversion data showed that in the reservoirs already found by the OA well, the impedance attribute could be directly correlated to hydrocarbon presence as was the case in Reservoirs A, B and C that had thicknesses in excess of 190 feet. Integration of the parameters from the inversion with six other seismic attributes in a back propagation neural networks (BPNN) scheme resulted in the identification of five exploration prospects (A, B, C, D and E) with different characters. Prospects A and C were identified as flat spots; with very good hydrocarbons presence. Prospect E was a possible anticlinal structure which extended beyond the limits of the study area and would require additional data to completely evaluate its potential. Prospects B and D were stratigraphic targets which would require two wells to be drilled to rest them. The study concluded that the integration of pre-stack seismic inversion and artificial neural networks enhanced the exploration potentials of the “OA”-Field in terms of prospect delineation which could eventually lead to identification of additional hydrocarbon reserves and resources.