Graphics Reference
In-Depth Information
Input
Plant Output
×10 5
2.2
180
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2 0
160
140
120
100
80
60
40
20
100
200
300
400
500
600
700
0 00
200 00
400 00
600 00
Error
NN Output
2
180
160
140
120
100
80
60
40
220
1.5
1
0.5
0
-0.5
-1 0
100
200
300
400
500
600
700
0 00
200 00 400
Time (s)
500 00
700
Time (s)
FIGURE 4.8
Data collected from Experiment 1.
over the test period. The mean difference between the frame rates generated by the
neural network model and the actual application was 0.00455.
In the same order, the graphs for the second experiment using the game are shown
in Figure  4.9. The neural network was able to model closely the characteristics of
the rendering process in the second application with a mean difference of 0.00896
in frame rate. All networks were trained using the Levenberg-Marquardt algorithm
over 200 epochs for over 5,000 frame samples.
4.4.2
anfis m odel
In Experiment 3, 120,000 input and output data pairs, each consisting of a vertex
count and frame rate, were collected. Figure 4.10 is a screenshot of the 3D rendering
application. Eighty thousand data pairs were used for training the ANFIS and the
remaining data pairs for validation. We used the ANFIS tool from the MATLAB
Fuzzy Logic Toolbox for the design and training of the fuzzy inference system.
The ANFIS model output was compared with the user-defined reference out-
put in Figure  4.11. We can observe from the figure that the output of the ANFIS
model closely follows the reference output. The error over the entire duration of the
Search WWH ::




Custom Search