Total organic carbon estimation in shale-gas reservoirs using seismic genetic inversion with an example from the Barnett Shale

摘要

Our goal is to invert total organic carbon (TOC) using the multilayer perceptron neural network. Three-dimensional seismic data recorded near two horizontal wells drilled in the lower Barnett Shale Formation are used as input to train a neural machine, while the calculated acoustic impedance from sonic and density well-log data for the horizontal wells is used as output. The multilayer perceptron neural machine is trained in a supervised mode, and weights of connections are calculated. The full 3D seismic data are then propagated through this machine, and a cube of acoustic impedance is inverted. A crossplot of the acoustic impedance versus the TOC is used to provide a linear relationship between these two parameters. This relationship is used to suggest a 3D TOC cube from the inverted cube of the acoustic impedance. Obtained results are checked with the Schmoker's TOC of another horizontal well drilled in the lower Barnett. These results show the ability of the genetic inversion to enhance characterization of shale-gas reservoirs.