Page 130 - Machine Learning for Subsurface Characterization

Stacked neural network architecture Chapter 4 107

FIG. 4.1 Flowchart of the two-step method, also referred as the SNN model, involving nine neural
network models for dielectric dispersion (DD) log synthesis. In the first step, 1 ANN model with 15
inputs and 8 outputs is trained to simultaneously synthesize the 8 DD logs; following that, each of the
8 DD logs is assigned a rank based on the prediction accuracy achieved by the ANN model during
the simultaneous synthesis. In the second step, eight distinct ANN models are trained one at a time to
sequentially synthesize one of the eight DD logs based on the rank assigned in the first step, starting
by predicting the DD log that was synthesized with the highest prediction accuracy and ending by
predicting the DD log that was synthesized with the lowest prediction accuracy.
the 8 DD logs are the targets (outputs). The proposed stacked neural network
(SNN) architecture combines nine separate artificial neural network (ANN)
models with fully connected layers to synthesize the 8 DD logs in two steps.
In the first step (Fig 4.1, top), one ANN model with 15 inputs and 8 outputs
is trained to simultaneously synthesize the 8 DD logs; following that, each of
the eight DD logs is assigned a rank based on the prediction accuracy achieved
by the ANN model during the simultaneous synthesis. In the second step
(Fig. 4.1, bottom), eight distinct ANN models are trained one at a time to
sequentially synthesize one of the eight DD logs based on the rank assigned
in the first step, starting with the DD log that was synthesized with the highest
prediction accuracy and ending with the DD log that was synthesized with the
lowest prediction accuracy. When sequentially training the 8 ANN models
implemented in the second step for the sequential synthesis of the 8 DD logs
(Fig. 4.1), the ith ANN model that synthesizes the ith ranked DD log is fed with
all the previously predicted or measured higher-ranked DD logs (1 to i 1) and
the 15 conventional logs. In other words, the ith ANN model in the SNN model
has 14+i inputs and 1 output, where i ranges from 1 to 9.
Various stacked neural network architectures can be designed by changing
the number and arrangement of ANN models in the stack, the connection
between pairs of ANN models, and the architecture of each ANN model in
the stack. In general, the architecture of an ANN model is defined by the number
of hidden layers, the number of neurons in a hidden layer, the number and type
of inputs and outputs, and the connections between layers and those between the
neurons. Similarly, the architecture of a stacked neural network (SNN) model is
defined by the number of ANN models in the stack, the number and type of
inputs and outputs of each ANN model in the stack, and the connections
between the ANN models in the stack.