3D geomechanical modeling in complex environments using the neural network

Modern computational tools and Machine Learning algorithms allowed automatic identification of petrophysical and lithological properties using geophysical profiles and seismic data as training set. Once knowing the input data limitation and established the parameters, it is possible to create in an agile and efficient way 1D and 3D models to determine petrophysical, mechanical and geological properties. Our goal is to apply neural networks to develop a 3D goemechanical model of a complex portion of the evaporitic section of a Brazilian offshore field to be used in well drilling. The modeled properties were lithofacies, Dtc (compressional transit time), Dts (shear transit time), density, Young’s Modulus and Poisson’s Ratio. The mechanical, physical and geological 3D models were distributed using neural networks techniques while the input data were property cubes produced by seismic inversion. The 3D models were considered adequate because they were able to identify the heterogeneity of the different saline lithotypes despite the area’s complexity. The quantitative validation were also adequate, with errors rate under 5%.