Precise geopressure predictions in active foreland basins: An application of deep feedforward neural networks

Precise geopressure predictions in younger foreland basins of the world is a challenging job. Well based pressure prediction gives only one-dimensional information about the pressure variations. Pressure predictions at a basin level by generating two-dimensional pressure sections requires integration of seismic and well data using geostatistical analysis. A combination of linear (multiattribute analysis) and non-linear (neural network) geostatistical analysis for precise two-dimensional pressure prediction is applied in an active foreland basin. Potwar Basin, which is the major oil and gas producing basin of Pakistan, exhibits abnormally high pressures within Neogene sediments even at shallow depths. Deep Feedforward Neural Network (DFNN) analysis is applied for the first time to predict pore pressures on 2D seismic transects of multiple producing fields of Potwar Basin for identification of abnormal pressure intervals within Murree Formation. Impedance sections along with original seismic trace and pressure curve from well are used as an input to train the seismic attributes to populate the pore pressure laterally. A total of 20 attributes are selected for data training using seismic based multiattribute analysis resulting in more than 85 % correlation in all the wells with a maximum correlation value of 91 %. Furthermore, DFNN analysis is then applied to enhance the results which showed more than 97 % coefficient in Qazian-1X, Balkassar OXY-01 and Amirpur-01 wells. DFNN analysis for demarcation of abnormal pressure zones depicts pressure results with an average 92 % correlation coefficient and provided a constrained pore pressure prediction workflow for getting consistent and reliable results in structurally complex foreland basins.