28.02.2014 Views

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

The Development of Neural Network Based System Identification ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

132 CHAPTER 5 NN BASED SYSTEM IDENTIFICATION: RESULTS AND DISCUSSION<br />

using the upper and lower confidence intervals (CI) limit <strong>of</strong> the optimal weights. <strong>The</strong><br />

measure <strong>of</strong> upper U i and lower L i limit width <strong>of</strong> the network output is given by taking<br />

the average <strong>of</strong> interval range over the measurement sample [Khosravi et al., 2011]. This<br />

is also known as Normalised Mean CI width (NMCIW) and can be used to show the<br />

variation <strong>of</strong> the targets. <strong>The</strong> measure formulation is given as follows:<br />

NMCIW = 1 N<br />

∑NR<br />

(U i − L i ) (5.1)<br />

i=1<br />

<strong>The</strong> average range <strong>of</strong> the upper and lower output performance (NMCIW) for each<br />

noise level cases are given in Table 5.3. As can be seen from the result, the weights<br />

set added random noise with s = 0.2 provide a wide range <strong>of</strong> confidence interval<br />

(28.3546%) compared with the range <strong>of</strong> the measurement between −1 rad/s to 1 rad/s.<br />

This indicates that the prediction is imprecise and unreliable to produce predictions<br />

that represent the real target values. In this study, a 20% threshold value is used to<br />

determine whether the CI is too wide or not.<br />

Table 5.3 <strong>The</strong> average RMSE for various noise levels applied to optimum weights <strong>of</strong> MLP network (4<br />

hidden neurons with 3 past outputs and 1 past input).<br />

Standard Deviation<br />

<strong>of</strong> Noise<br />

0.01<br />

0.05<br />

0.1<br />

0.2<br />

Test Error Statistics<br />

<strong>System</strong><br />

Responses<br />

RMSE RMSE (%) R2 NMCIW (%)<br />

p 0.0375 8.9256 0.9919 0.4232<br />

q 0.0082 2.5568 0.9992 1.4598<br />

p 0.0391 9.3019 0.9913 2.0441<br />

q 0.0134 4.1912 0.9980 7.2532<br />

p 0.0683 16.2613 0.9733 4.4477<br />

q 0.0365 11.3787 0.9851 14.7242<br />

p 0.1779 42.3673 0.8185 8.8449<br />

q 0.0677 21.1145 0.9488 28.3546

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!