Simulation of the reactor-regenerator-main fractionator fluid catalytic cracking unit using artificial neural networks

Vasile Mircea Cristea, Raluca Roman, Paul Şerban Agachi

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

The present work it is a successful approach for modelling the dynamic behaviour of the FCC unit, using Artificial Neural Networks (ANN). An analytical model, validated with construction and operation data, has been used to produce a comprehensive input-target set of training data. The novelty of the model consists in that besides the complex dynamics of the reactor-regenerator system, it also includes the dynamic model of the main fractionator. A new five-lump kinetic model for the riser is also included. Consequently, it is able to predict the final production rate of the main products, gasoline and diesel. The architecture and training algorithm used by the ANN are efficient and this is proved by the results obtained both on training set and set of input-target data not met during the training procedure. The same good ANN performance has been obtained by the comparison between dynamic simulations results emerged from the ANN model versus first principle modelling, both using the same randomly varying inputs. The computation time is considerably reduced when using the ANN model, compared to the use of the analytical model. The presented results show the incentives and benefits for further exploiting the ANN model as internal model for Model Predictive Control industrial implementation.

Original languageEnglish
Pages (from-to)125-132
Number of pages8
JournalStudia Universitatis Babes-Bolyai Chemia
Volume1
Publication statusPublished - 2009

Fingerprint

Fluid catalytic cracking
Regenerators
Neural networks
Analytical models
Model predictive control
Network performance
Gasoline
Dynamic models
Kinetics
Computer simulation

All Science Journal Classification (ASJC) codes

  • Chemistry(all)

Cite this

@article{e429c8e219f34d4d8b50ab8e21c6afbc,
title = "Simulation of the reactor-regenerator-main fractionator fluid catalytic cracking unit using artificial neural networks",
abstract = "The present work it is a successful approach for modelling the dynamic behaviour of the FCC unit, using Artificial Neural Networks (ANN). An analytical model, validated with construction and operation data, has been used to produce a comprehensive input-target set of training data. The novelty of the model consists in that besides the complex dynamics of the reactor-regenerator system, it also includes the dynamic model of the main fractionator. A new five-lump kinetic model for the riser is also included. Consequently, it is able to predict the final production rate of the main products, gasoline and diesel. The architecture and training algorithm used by the ANN are efficient and this is proved by the results obtained both on training set and set of input-target data not met during the training procedure. The same good ANN performance has been obtained by the comparison between dynamic simulations results emerged from the ANN model versus first principle modelling, both using the same randomly varying inputs. The computation time is considerably reduced when using the ANN model, compared to the use of the analytical model. The presented results show the incentives and benefits for further exploiting the ANN model as internal model for Model Predictive Control industrial implementation.",
author = "Cristea, {Vasile Mircea} and Raluca Roman and Agachi, {Paul Şerban}",
year = "2009",
language = "English",
volume = "1",
pages = "125--132",
journal = "Studia Universitatis Babes-Bolyai Chemia",
issn = "1224-7154",
publisher = "Universitatea Babes-Bolyai, Catedra de Filosofie Sistematica",

}

Simulation of the reactor-regenerator-main fractionator fluid catalytic cracking unit using artificial neural networks. / Cristea, Vasile Mircea; Roman, Raluca; Agachi, Paul Şerban.

In: Studia Universitatis Babes-Bolyai Chemia, Vol. 1, 2009, p. 125-132.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Simulation of the reactor-regenerator-main fractionator fluid catalytic cracking unit using artificial neural networks

AU - Cristea, Vasile Mircea

AU - Roman, Raluca

AU - Agachi, Paul Şerban

PY - 2009

Y1 - 2009

N2 - The present work it is a successful approach for modelling the dynamic behaviour of the FCC unit, using Artificial Neural Networks (ANN). An analytical model, validated with construction and operation data, has been used to produce a comprehensive input-target set of training data. The novelty of the model consists in that besides the complex dynamics of the reactor-regenerator system, it also includes the dynamic model of the main fractionator. A new five-lump kinetic model for the riser is also included. Consequently, it is able to predict the final production rate of the main products, gasoline and diesel. The architecture and training algorithm used by the ANN are efficient and this is proved by the results obtained both on training set and set of input-target data not met during the training procedure. The same good ANN performance has been obtained by the comparison between dynamic simulations results emerged from the ANN model versus first principle modelling, both using the same randomly varying inputs. The computation time is considerably reduced when using the ANN model, compared to the use of the analytical model. The presented results show the incentives and benefits for further exploiting the ANN model as internal model for Model Predictive Control industrial implementation.

AB - The present work it is a successful approach for modelling the dynamic behaviour of the FCC unit, using Artificial Neural Networks (ANN). An analytical model, validated with construction and operation data, has been used to produce a comprehensive input-target set of training data. The novelty of the model consists in that besides the complex dynamics of the reactor-regenerator system, it also includes the dynamic model of the main fractionator. A new five-lump kinetic model for the riser is also included. Consequently, it is able to predict the final production rate of the main products, gasoline and diesel. The architecture and training algorithm used by the ANN are efficient and this is proved by the results obtained both on training set and set of input-target data not met during the training procedure. The same good ANN performance has been obtained by the comparison between dynamic simulations results emerged from the ANN model versus first principle modelling, both using the same randomly varying inputs. The computation time is considerably reduced when using the ANN model, compared to the use of the analytical model. The presented results show the incentives and benefits for further exploiting the ANN model as internal model for Model Predictive Control industrial implementation.

UR - http://www.scopus.com/inward/record.url?scp=70349223736&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70349223736&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:70349223736

VL - 1

SP - 125

EP - 132

JO - Studia Universitatis Babes-Bolyai Chemia

JF - Studia Universitatis Babes-Bolyai Chemia

SN - 1224-7154

ER -