Online system identification development based on recursive weighted least square neural networks of nonlinear hammerstein and wiener models.
The realistic dynamics mathematical model of a system is very important for analyzing a system. The mathematical system model can be derived by applying physical, thermodynamic, and chemistry laws. But this method has some drawbacks, among which is difficult for complex systems, sometimes is untr...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English English |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/8404/1/24p%20AYAD%20MAHMOOD%20KWAD.pdf http://eprints.uthm.edu.my/8404/2/AYAD%20MAHMOOD%20KWAD%20COPYRIGHT%20DECLARATION.pdf http://eprints.uthm.edu.my/8404/3/AYAD%20MAHMOOD%20KWAD%20WATERMARK.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The realistic dynamics mathematical model of a system is very important for analyzing
a system. The mathematical system model can be derived by applying physical,
thermodynamic, and chemistry laws. But this method has some drawbacks, among
which is difficult for complex systems, sometimes is untraceable for nonlinear behavior
that almost all systems have in the real world, and requires much knowledge. Another
method is system identification which is also called experimental modeling. System
identification can be made offline, but this method has a disadvantage because the
features of a dynamic system may change over time. The parameters may vary as
environmental conditions change. It requires big data and consumes a long time. This
research introduces a developed method for online system identification based on the
Hammerstein and Wiener nonlinear block-oriented structure with the artificial neural
networks (NN) advantages and recursive weighted least squares algorithm for optimizing
neural network learning in real-time. The proposed method aimed to obtain a maximally
informative mathematical model that can describe the actual dynamic behaviors of a
system, using the DC motor as a case study. The goodness of fit validation based on
the normalized root-mean-square error (NRMSE) and normalized mean square error, and
Theil’s inequality coefficient are used to evaluate the performance of models. Based on
experimental results, for best Wiener parallel NN model and series-parallel NN model
are 93.7% and 89.48%, respectively. Best Hammerstein parallel NN polynomial based
model and series-parallel NN polynomial model are 88.75% and 93.9% respectively,
for best Hammerstein parallel NN sigmoid based model and series-parallel NN sigmoid
based model 78.26% and 95.95% respectively, and for best Hammerstein parallel NN
hyperbolic tangent based model and series-parallel NN hyperbolic tangent based model
70.7% and 96.4% respectively. The best model of the developed method outperformed the
conventional NARX and NARMAX methods best model by 3.26% in terms of NRMSE
goodness of fit. |
|---|