Page 265 - Machine Learning for Subsurface Characterization

Shallow and deep machine learning models Chapter 8 229

FIG. 8.3 Comparisons of measured NMR T2 distributions against those synthesized by processing
raw logs in the testing dataset using (A) OLS, (B) LASSO, (C) ElasticNet, (D) SVR, (E) kNNR, and
(F) ANN models.

reconstructing the NMR T2 distribution; following that, the learned abstractions
and features of NMR T2 are related to the easy-to-acquire logs for the synthesis
of NMR T2 distribution. The three models are variational autoencoder assisted
neural network (VAE-NN), variational autoencoder with convolutional layer
assisted neural network (VAEc-NN), and generative adversarial network
assisted neural network (GAN-NN). These three models adopt a two-step
training process. In the first step, VAE-NN and VAEc-NN models build a
decoder network and the GAN-NN model builds a generator network for
purposes of learning to reconstruct NMR T2 distribution. In the second step,
a simple ANN with 3–5 hidden layers is connected to the trained decoder/
generator for learning the complex hidden relationship between the “easy-to-