Alomair
PTE586 Anas HW#3
09/16/2010
1.
Designing the network will begin with setting the network architecture. The
network architecture establishes how the network weights are connected. Then, a
learning algorithm such as back propagation should be established which
automatically adjust the weights to get the desired data. The network consists of
input layer, hidden layer, and output. Then, the activation rule should be obtained
which is a transfer function, which combines the input with weight layers to get net
input. A sigmoid function can be used as a transfer function.
Training the network: most of the data are used to train the network and the rest
are used to test it, e.g. 80% are used for training and 20% are used to test the
network. In the training, the network model uses most of the input parameters and
try architectures, different learning and activation rules. Then, the optimum
network is chosen by testing each trial with the testing data and getting the most
suitable results.
2.
The graph of the network is attached in scanned pdf document.
3.
The statistical method used before applying ANN is Normalization.
Normalization is used before applying ANN to alter the range of input values to get
more accurate results. When large and small values are used without normalization,
the results sometimes could be not accurate because of the magnitude effect.
4.
Fuzzy logic can be used with neural network to account for the fuzzy qualifiers.
In the case of question 4, the hourly water rate and the hourly temperature can be
used as input. And then, Fuzzy N-dimension is obtained. The output should be gfr
(gas flow rate)