Information Technology Reference
In-Depth Information
Here, extreme learning machine method is used to identify the parameters of
the proposed QNN. Assume that the QNN can approximate the
N
samples with
zero error which means that
j
=1
o
j
−
y
j
= 0. Thus, there exist
β
i
,
w
i
and
b
i
such that
N
β
i
sin
2
[
π
2
(
w
i
·
x
j
+
b
i
)] =
y
j
(
j
=1
,
2
,...,N
)
(9)
i
=1
The above
N
equations can be written compactly as
H
β
=
Y
,where
H
(
w
1
,...,
w
N
,b
1
,...,b
N
,
x
1
,...,
x
N
)
⎡
⎣
⎤
⎦
sin
2
[
2
(
w
1
·
sin
2
[
2
(
w
N
·
x
1
+
b
1
)]
···
x
1
+
b
N
)]
.
.
sin
2
[
2
(
w
1
·
x
N
+
b
1
)]
···
sin
2
[
2
(
w
N
·
x
N
+
b
N
)]
(10)
=
···
N
N
×
After the input weights
w
i
and thresholds
b
i
of hidden layer nodes are chosen
arbitrarily, the QNN can be simply considered as a linear system. The output
weights
β
of QNN can be analytically determined according to Equation (11).
⎡
⎣
⎤
⎦
⎡
⎣
⎤
⎦
†
sin
2
[
2
(
w
1
·
sin
2
[
2
(
w
N
·
x
1
+
b
1
)]
···
x
1
+
b
N
)]
y
1
.
y
N
.
.
sin
2
[
2
(
w
1
·
x
N
+
b
1
)]
β
=
(11)
···
sin
2
[
2
(
w
N
·
x
N
+
b
N
)]
···
N
N×
N×m
4 Results and Discussion
In the Texaco gasification process, the syngas components such as CO, H
2
and
CO
2
are very critical for instructing the regular operations. However, in an appli-
cation case of fertilizer plant, the syngas components are calculated through ex-
periment analysis. Off-line manual computing is time-delay, and tends to greatly
reduce the operational eciency. Therefore, soft computing technique based on
the proposed ELM-QNN method is adopted to execute the online assessment
of the syngas components. Some measurable process variables are used to in-
directly calculate the key unmeasurable variables, including CO, H
2
and CO
2
concentration in the Texaco syngas. The measurable variables in the gasification
process include the characteristics of coal, slurry, oxygen, quenching water, etc.
After smoothed and normalized, 253 groups of sample data are exposed to the
ELM-QNN. Among the sample data, 200 groups are the training data for iden-
tifying the parameters of QNN, and the remaining are the testing data which
are used to validate the generalization capability of ELM-QNN.
Though each of the measurable variables can reflect the process information
partly, there exist correlativities between some of them. In order to eliminate the
redundant information stored in variables, principal component analysis method
Search WWH ::
Custom Search