Environmental Engineering Reference
In-Depth Information
Table 4
Evaluation of eco-innovation implementation
Indicator
Respondent
Strongly
disagree
Disagree
Neither
agree
nor
disagree
Agree
Strongly
agree
Score-
team
member
Score-
indicator
1
2
3
4
5
1
0.3
0.7
3.7
3.875
Our unit
management
often uses
novel systems
to manage
eco-innovation
1
4
2
0.4
0.6
3.6
3
0.8
0.2
4.2
4
…
…
…
…
1
0.5
0.5
2.5
2.975
Our unit
often
emphasizes
developing
new
eco-products
through new
technologies
to use as
little energy as
possible
0.8
0.2
3.2
2
0.7
0.3
3.3
3
0.1
0.9
2.9
4
5.2 Identification of Relationships
In this research, a multilayer feed-forward neural network has been applied by the
use of the back-propagation algorithm, and optimisation weights according to the
Levenberg-Marquardt algorithm (LM), gradient descent momentum and an adap-
tive learning rate (GDX) algorithm. The neural network structure has been deter-
mined in an experimental way, by the comparison of learning and testing sets for
the different number of layers and hidden neurons. RMSE have been calculated as
the average of 20 iterations for each structure of neural network with a number to
the extent of 20 hidden neurons.
In order to eliminate the overtraining of ANN (too strict function adjustment
to data) and to increase the estimation quality, the data set has been divided into
learning (P1-P62) and testing sets (P63-P77). The learning process of ANN aims
to seek the minimal error in the testing set. The results as the root mean square
errors (RMSE) for the learning and testing sets are presented in Table
5
. The
results for ANNs are compared with the linear model calculating according to the
ordinary least squares method.
The results presented in Table
5
indicate that the least error in the testing set
was generated with the use of ANN trained according to the GDX algorithm.
The least RMSE in the learning set was calculated according to the LM algorithm
Search WWH ::
Custom Search