Neuro fuzzy modeling of propylene polymerization
In the present study, a neuro fuzzy kinetic model was developed to predict production rate for bulk homo-polymerization of propylene in an industrial loop reactors. The adaptive-network-based fuzzy inference system (ANFIS) technique was trained with recorded data and generated the membership functio...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2011
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/26873/1/LeilaEzzatzadeganMFKKSA2011.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In the present study, a neuro fuzzy kinetic model was developed to predict production rate for bulk homo-polymerization of propylene in an industrial loop reactors. The adaptive-network-based fuzzy inference system (ANFIS) technique was trained with recorded data and generated the membership function and rules which most excellent expounded the input/output correlations in the process. Three adaptive network- based fuzzy inference systems were presented. The three neuro fuzzy systems are ANFIS based grid partitioning, ANFIS based subtractive clustering, and ANFIS based Fuzzy C-means clustering. For implementation of the resent technique the MATLAB (2010a) codes were efficiently employed. The effect of different parameters for training the model was studied and the performances of consequential FIS were compared. A real-world homo-polymerization production rate data set was gathered from a typical petrochemical complex and after pre-treating was used for training of ANFIS. ANFIS model was generated and tested using training and testing data from that data set. The performance of best obtained network was checked by its generalization ability in predicting 30% of the unseen data. Excellent prediction with Root Mean Square Error (RMSE) of 0.0096 was observed. ANFIS based subtractive clustering outperformed ANFIS based grid partitioning and ANFIS based Fuzzy C-means clustering due to its fitness in the target problem. At the next step, the result of best ANFIS model was compared with a first principal model and then this model was modified and its result was also compared with ANFIS model. This paper shows the appropriateness and superiority of ANFIS for the quantitative modeling of production rate than first principal modeling. |
|---|