Shale Anisotropic Parameters Predicting Based on Deep Neural Networks

In shale reservoirs, anisotropy cannot be ignored. However, in conventional well logging measurements, shale anisotropic parameters cannot be measured directly. We build a deep neural network to estimate the nonlinear relationship between the well log curves and the anisotropic parameters. In the composition of the training set, we use the results calculated from the anisotropic self-consistent approximation + differential effective medium model (SCA + DEM) model based on equivalent medium theory and Chapman model based on wave-induced exchange of fluids between pores and cracks. We hope to get a neural network that describe both the complex composition of shale and the fluid effects. Finally, the trained neural network is applied to a field data example. The results show that the anisotropic parameters estimated by the neural network agree well with the real values.