Machine learning and seismic attributes for prospect identification and risking: an example from Offshore Australia

The consistently increasing number of seismic attributes makes the use of every seismic attribute difficult if not impossible. Machine learning algorithms demonstrated a good capability of selecting useful attributes and extracting the maximum of their hidden features. In this study, seismic attributes are used to detect hydrocarbon-saturated reservoirs from Poseidon field, Offshore Australia. Feedforward Artificial Neural Networks (ANN) are implemented to combine seismic attributes and convert them to Gas chimney probability cube, and to Gamma Ray cube. Next, pre-trained Convolutional Neural Network (CNN) is trained using a selected attribute set to detect subtle faults that hydrocarbon might have used to migrate from the source rock to the trap. The integration of fluid-related attributes such as Scaled Poisson Reflectivity (SPR) and AVO Gradient, the predicted Gamma Ray, and the high density detected faults combined with the gas chimneys probability cube helped find new prospects in the area. They also helped assess the risk of the prospects and order them in terms of their associated geological risk.