The value of using seismic velocities for pore-pressure prediction highly depends on resolution and accuracy of obtained estimates. In zero-offset VSP scenario, seismic velocity ahead of the bit is typically estimated from acoustic impedance using Gardner’s relation. The prediction error highly depends on the error in Gardner’s regression estimated from well-logs. We developed a data-driven (MLbased) methodology for prediction of velocity and density head of the bit based on ML training on the available log and seismic data for a single well and a single shot location. Using an experimental well of Aramco in Houston, we demonstrate that the new method can improve upon conventional approach.
Keywords:reservoir characterization, artificial intelligence, impedance, upstream oil & gas, machine learning, neural network, seismic data, information, wavefield, workflow
This content is only available via PDF.
Copyright 2021, Society of Exploration Geophysicists