Page 269 - PVT Property Correlations

Artificial Neural Network Models for PVT Properties Chapter | 10 235

FIGURE 10.6 ANN structure for
W1 W5
I1 H1 O1 Problem 1. ANN, artificial neural
W2 W6 network.

W3
W7
I2 H2 O2
W4 W8

B1 B2

TABLE 10.2 Training Record for the ANN in Problem 1

Input Nodes Output Nodes Bias Nodes
I1: 0.03 O1: 0.01 B1: 0.3
I2: 0.20 O2: 0.99 B2: 0.7

TABLE 10.3 Weight Initialization Values for the ANN in Problem 1
Hidden Layer Weights Output Layer Weights
W1: 0.12 W3: 0.20 W5: 0.40 W7: 0.55
W2: 0.15 W4: 0.40 W6: 0.20 W8: 0.50

Table 10.2. Use learning rate of 0.6. Perform the calculations to train the
ANN model.

Solution of Problem 1
The calculation process of the ANN consists of four steps: (1) to initialize
the network weights (assign values for the weights connecting the input layer
nodes to the hidden layer nodes and the weights connecting the hidden layer
nodes to the output nodes); (2) to perform the feed-forward calculations; (3)
to calculate the network error; and (4) to perform the feed-backward calcula-
tions. Steps 2 4 are repeated until the calculated error satisfies the network
convergence criteria. Table 10.3 provides the initial weights for both the hid-
den layer and output layer nodes.