How AI-derived well data helps in the estimation of pore-pressure from seismic data
The Powder River Basin is known to contain anomalous pressure cells, especially within the Muddy Sandstone. It is desirable to map these areas prior to drilling and seismic data is the best, if not the only, reconnaissance data type available. Overpressure in unconventional on-shore basins is complicated by complex geology, low permeability and the presence of TOC in the shale reservoirs. Combining well information with seismic elastic pre-stack attributes allows us to estimate higher pressured zones that can be beneficial for production and can help avoiding negative drilling surprises.
Seismic inputs for such a study are 3D estimated volumes of the elastic seismic attributes Vp and density. Both data types are derived from elastic pre-stack inversions, a process which requires careful, amplitude-preserved and phase stabilized seismic processing. The pore-pressure model is constructed using direct pressure information from drilling such as Diagnostic Fracture Initiation Tests (DFIT), Drill-Stem Tests (DST), mud weight and all other information that indicates pressure changes. A cross-plot of the pressure values (expressed in terms of vertical effective stress) and well-log measured velocities allows a regression based on the Bower’s equation effectively linking Vp to pore pressure.
Often, drilling info is available at a certain well but no sonic log was recorded. A novel approach using machine learning was applied to predict GR, neutron, resistivity, density and sonic logs at well locations that have at least one of these logs. The prediction is performed using a guided decision tree method and additional estimated sonic logs allow for a more accurate estimation of the Bower’s variables. If both of those data types (i.e., pressure information and sonics) are available at the same well, a higher confidence in the free parameter estimation is achieved. By applying the regression relationship resulting from the cross-plotting to the seismic-based Vp volume, the actual pressure can be estimated away from the wells at all points in x, y, z.
The seismically derived Vp values need to be corrected for TOC. We use the density attribute to estimate TOC at every x, y and z location to correct the seismic Vp volume. After the correction was performed, the Bower’s equation is applied to the corrected dataset, which produces a vertical effective stress cube. The next step uses the Terzaghi equation to perform the conversion from vertical effective stress to pore-pressure.
A validation of the seismic pore-pressure values with normalized production information and gas readings shows a high degree of certainty. The method should be applicable in the tight rock environment in all on-shore basins and should yield similar results.
AAPG Datapages/Search and Discovery Article #90374 © 2020 AAPG Rocky Mountain Section Meeting, 2020 Vision: Turn Hindsight to Foresight, Grand Junction, Colorado, September 13-15, 2020 (CANCELLED)