Geoscience Reference
In-Depth Information
600
BPLM Rel
RNN Rel
Demand
400
200
0
Month
Fig. 4.
Comparison of performance by different methods of ANN.
model. The validation results are shown in Fig. 4. It is important to note
that the validation data should not have been used as part of the training
process in any capacity. The results are discussed in the following section.
4. Results and Discussion
The applicability of different kinds of neural networks for the probabilistic
analysis modeling for random variables are summarized. The comparison
comprehends two network algorithms (multilayer neural networks). This is
a relevant result for neural networks learning because most of the practical
applications employ the neural learning heuristic back-propagation, which
uses different activation functions. Back-Propagation using Levenberg-
Marguardt algorithm needs no additional executable program modules in
the source code. But RNN has taken less epoch number compared to that
of ANN. In addition, the number of epochs to find the optimal solution at
different tests is significantly reducing in RNN. Further, the figures showed
a relatively small running time taken to find the optimal solution when
an RNN replaces the ANN model. This is exhibited in Fig. 4. The release
from ANN, RNN and Actual practice are plotted for testing the perfor-
mance of classical linear programming and heuristic method as applied to
the problem of reservoir operation.
