Artificial neural network modeling and performance optimization on biodiesel production process

Research output: Contribution to journalArticle

Abstract

In meeting with the challenge of energy crisis, biodiesel emerges as an important renewable and environmental-friendly fuel to support the world demand. To enable the industrial process to be more efficient, understanding on the process characteristic of biodiesel production through modeling is crucial for process optimization and control. Artificial Neural Network (ANN) was used to predict the dynamic trending for 10 reaction components based on the input parameter of time. The ANN used is feed-forward type with single hidden layer and coupled with Levenberg-Marquardt (LM) training algorithm. To illustrate the capability of ANN, the case study from literature with base-catalyzed transesterification on Jatropha curcas-waste food oil mixture with potassium hydroxide as catalyst is presented. The input neuron represents time and the output neurons are concentrations of all 10 reaction components inclusive of catalyst, free fatty acid and soap. The trained 3-layer ANN had shown the most satisfied performance with low number of hidden neurons. No significant improvement noted from the changes of initial range of weights and biases. The newly proposed updating method for damping factor in LM algorithm had reduced the number of epochs and thus improved the simulation time against the default method. Setting the updating factor of damping factor to be higher than unity is noted to contribute to lower number of epochs. The optimized ANN has shown great performance and potential for the application on real process modeling and control in biodiesel production process.

Original languageEnglish
Pages (from-to)4854-4864
Number of pages11
JournalJournal of Applied Sciences Research
Volume8
Issue number10
Publication statusPublished - 2012

Fingerprint

Biodiesel
Neural networks
Neurons
Damping
Potassium hydroxide
Catalysts
Soaps (detergents)
Transesterification
Fatty acids
Process control

Keywords

  • Correlation coefficient
  • Levenberg-Marquardt algorithm
  • Mean squared error
  • Optimization
  • Transesterification

ASJC Scopus subject areas

  • General

Cite this

@article{418aa50fcf68487d80b7c784a0671e2e,
title = "Artificial neural network modeling and performance optimization on biodiesel production process",
abstract = "In meeting with the challenge of energy crisis, biodiesel emerges as an important renewable and environmental-friendly fuel to support the world demand. To enable the industrial process to be more efficient, understanding on the process characteristic of biodiesel production through modeling is crucial for process optimization and control. Artificial Neural Network (ANN) was used to predict the dynamic trending for 10 reaction components based on the input parameter of time. The ANN used is feed-forward type with single hidden layer and coupled with Levenberg-Marquardt (LM) training algorithm. To illustrate the capability of ANN, the case study from literature with base-catalyzed transesterification on Jatropha curcas-waste food oil mixture with potassium hydroxide as catalyst is presented. The input neuron represents time and the output neurons are concentrations of all 10 reaction components inclusive of catalyst, free fatty acid and soap. The trained 3-layer ANN had shown the most satisfied performance with low number of hidden neurons. No significant improvement noted from the changes of initial range of weights and biases. The newly proposed updating method for damping factor in LM algorithm had reduced the number of epochs and thus improved the simulation time against the default method. Setting the updating factor of damping factor to be higher than unity is noted to contribute to lower number of epochs. The optimized ANN has shown great performance and potential for the application on real process modeling and control in biodiesel production process.",
keywords = "Correlation coefficient, Levenberg-Marquardt algorithm, Mean squared error, Optimization, Transesterification",
author = "Hui, {Liew Weng} and Zahira Yaakob and {Sheikh Abdullah}, {Siti Rozaimah}",
year = "2012",
language = "English",
volume = "8",
pages = "4854--4864",
journal = "Journal of Applied Sciences Research",
issn = "1816-157X",
publisher = "INSInet Publications",
number = "10",

}

TY - JOUR

T1 - Artificial neural network modeling and performance optimization on biodiesel production process

AU - Hui, Liew Weng

AU - Yaakob, Zahira

AU - Sheikh Abdullah, Siti Rozaimah

PY - 2012

Y1 - 2012

N2 - In meeting with the challenge of energy crisis, biodiesel emerges as an important renewable and environmental-friendly fuel to support the world demand. To enable the industrial process to be more efficient, understanding on the process characteristic of biodiesel production through modeling is crucial for process optimization and control. Artificial Neural Network (ANN) was used to predict the dynamic trending for 10 reaction components based on the input parameter of time. The ANN used is feed-forward type with single hidden layer and coupled with Levenberg-Marquardt (LM) training algorithm. To illustrate the capability of ANN, the case study from literature with base-catalyzed transesterification on Jatropha curcas-waste food oil mixture with potassium hydroxide as catalyst is presented. The input neuron represents time and the output neurons are concentrations of all 10 reaction components inclusive of catalyst, free fatty acid and soap. The trained 3-layer ANN had shown the most satisfied performance with low number of hidden neurons. No significant improvement noted from the changes of initial range of weights and biases. The newly proposed updating method for damping factor in LM algorithm had reduced the number of epochs and thus improved the simulation time against the default method. Setting the updating factor of damping factor to be higher than unity is noted to contribute to lower number of epochs. The optimized ANN has shown great performance and potential for the application on real process modeling and control in biodiesel production process.

AB - In meeting with the challenge of energy crisis, biodiesel emerges as an important renewable and environmental-friendly fuel to support the world demand. To enable the industrial process to be more efficient, understanding on the process characteristic of biodiesel production through modeling is crucial for process optimization and control. Artificial Neural Network (ANN) was used to predict the dynamic trending for 10 reaction components based on the input parameter of time. The ANN used is feed-forward type with single hidden layer and coupled with Levenberg-Marquardt (LM) training algorithm. To illustrate the capability of ANN, the case study from literature with base-catalyzed transesterification on Jatropha curcas-waste food oil mixture with potassium hydroxide as catalyst is presented. The input neuron represents time and the output neurons are concentrations of all 10 reaction components inclusive of catalyst, free fatty acid and soap. The trained 3-layer ANN had shown the most satisfied performance with low number of hidden neurons. No significant improvement noted from the changes of initial range of weights and biases. The newly proposed updating method for damping factor in LM algorithm had reduced the number of epochs and thus improved the simulation time against the default method. Setting the updating factor of damping factor to be higher than unity is noted to contribute to lower number of epochs. The optimized ANN has shown great performance and potential for the application on real process modeling and control in biodiesel production process.

KW - Correlation coefficient

KW - Levenberg-Marquardt algorithm

KW - Mean squared error

KW - Optimization

KW - Transesterification

UR - http://www.scopus.com/inward/record.url?scp=84872584259&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84872584259&partnerID=8YFLogxK

M3 - Article

VL - 8

SP - 4854

EP - 4864

JO - Journal of Applied Sciences Research

JF - Journal of Applied Sciences Research

SN - 1816-157X

IS - 10

ER -