Probabilistic neural network-based seismic attribute prediction of pore pressure: a case study in the Timimoun Basin, Algeria

This study aims to estimate the subsurface pore pressure (Pp) within the Frasnian shale in the Algerian Timimoun Basin, where there have been no prior studies. Seismic and well-log data were incorporated to assess the lateral distribution of Pp. A probabilistic neural network (PNN) model was implemented based on its capabilities to better integrate the diverse data sources and adapt to complex geological conditions. The model is trained using data from four wells, using the Pp log, and the composite traces of seismic inversion outputs, validated on one blind well and compared to a baseline method. Then, the model was extended from the boreholes to the entire study area. The application of the PNN algorithm exhibited a robust Pearson coefficient of correlation of 0.94 and 4 MPa as root-mean-square error, signifying a substantial alignment between the predicted and actual Pp values. This correlation emphasises that seismic data effectively encapsulates the physical implications of predicted volumes. The PNN-predicted Pp aligns with the structural patterns of the Frasnian shale in the subsurface, suggesting that the lateral distribution of Pp is influenced by these structural characteristics. The assessment demonstrates an abnormal subsurface Pp which poses a challenge to drilling operations for future well exploration or exploitation.