Neural Network Training

Architecture:

Feed-forward network with

• 35 input neurons—flow front [16] flow rates [3], permeability [16]
• 16 output neurons (flow front at the next time step)
• 15 hidden neurons in 1 layer

Training:

Using ~1500 parametric data sets from the process model

Testing:

~1500 test cases (different from the training sets)

Previous slide

Next slide

Back to first slide

View graphic version