IDENTIFIKASI NONLINIER PLANT AMONIAK MENGGUNAKAN METODA NEURO-FUZZY DAN WAVENET
The objective of the research was to apply the modeling technique based on measurements of input-output data of the system, called system identification. For the purpose of system identification, the neuro-fuzzy and wavenet (neural network adaptive wavelet) method was used. Both methods are very app...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/3165 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | The objective of the research was to apply the modeling technique based on measurements of input-output data of the system, called system identification. For the purpose of system identification, the neuro-fuzzy and wavenet (neural network adaptive wavelet) method was used. Both methods are very appropriate for nonlinear system, wile, in general, most system/plant in this world having nonlinearity characteristics. Neuro-fuzzy technique which is used in this research has ANFIS (Adaptive Neuro-Fuzzy Inference System) structure which applies least-square estimator and error back propagation method as a hybrid learning rule. On the other hand, back propagation algorithm and iterative minimization methods of gradient steepest descent is used as wavenet algorithm. Both of this methods were applied to identify the dynamic of ammonia synthesis loop and aqueous-ammonia binary distillation column unit of ammonia plant. Identification process was done using the real measurement data collected from PT. Kaltim Pasifik Amoniak, Bontang plant for synthesis loop unit and from PT. Petrokimia Gresik plant for binary distillation column. The result showed that both of the methods perform well in identification of nonlinear plant. The best model obtained from NH3 concentration approximated with sixth order model by neuro-fuzzy method. Comparing the result between neuro-fuzzy and wavenet, it is shown that the root mean square error (RMSE) of neuro-fuzzy method smaller than wavenet, but it needs longer computing time. Both methods have advantage and weakness caused by the difference in e.g. parameters initialization, network structure and learning rules algorithm. It is expected that obtaining the accurate model from this research can be further explored to design an appropriate controller in the future. |
|---|