- You are here:
-
-
All publications
- Determination de l'applicabilite des reseaux de neurons pour simuler le sechage du mais en systeme mixte
All publications from your selection
Determination de l'applicabilite des reseaux de neurons pour simuler le sechage du mais en systeme mixte
Authors: R. Lacroix, R. Kok et S. Fortin
Identifier:
Download file: https://library.csbe-scgab.ca/docs/journal/41/41_2_105_ocr.pdf
Published in: CBE Journal » CBE Journal Volume 41 (1999)
Download RAW file: https://library.csbe-scgab.ca/docs/journal/41/41_2_105_raw.pdf
DOI:
Description: This study sought to establish the applicability of artificial neural networks (ANNs) to the computer simulation oflow-temperature grain drying, and whether such a system could satisfactorily reproduce the results obtained with a procedural model. The procedural model, SIMMAIS, was used to generate the input and output data needed to train a back-propagation ANN,and as a basis for comparison. Based on the input data (harvest date, initial moisture content, air flow, height/diameter ratio of the silo, presence or absence of heating, and meteorological conditions) the model predicted the percentage ofgrain deterioration, the energy consumed in ventilation and heating, the endof- drying date, and the amounts of grain with moisture contents over 20% or over 18%. Nine trials were run with different numbers of processing elements in each hidden layer, learning rules (delta-rule and normalized cumulative delta-rule), transfer functions in the processing elements (sigmoid or hyperbolic tangent), learning cycles (100,000 to 500,000), and data for learning. The ANN showing the best performance was the one with 20 processing elements in the first hidden layer and lOin the second, 100,000 learning cycles, hyperbolic tangent transfer functions and the normalized cumulative delta learning rule. Overall, this ANN satisfactorily reproduced the results of the procedural model. Generally, the average relative deviation was between I%and 3%. There were, however, a few extreme cases where this deviation was close to 50% for the deterioration. For the other variables, the maximum relative error was between 20% and 25%. If the ANN is to be used for decision support, these extreme values will have to be taken into account.
Keywords:
Résumé:
Mots-clés:
Citation: R. Lacroix, R. Kok et S. Fortin 1999. DETERMINATION DE L'APPLICABILITE DES RESEAUX DE NEURONES POUR SIMULER LE SECHAGE DU MAis EN SYSTEME MIXTE. Canadian Agricultural Engineering 41(2):105-112.
Volume: 41
Issue: 2
Pages -
Contributor:
Date: 1999
Technical field:
Conference name:
Session name:
Other information:
Type: Text.Article
Format: PDF
Publication type:
Source:
Relation:
Coverage: Canada
Language 1: en
Language 2:
Rights: Canadian Society for Bioengineering
Notes:
Other files:
DOI:
Description: This study sought to establish the applicability of artificial neural networks (ANNs) to the computer simulation oflow-temperature grain drying, and whether such a system could satisfactorily reproduce the results obtained with a procedural model. The procedural model, SIMMAIS, was used to generate the input and output data needed to train a back-propagation ANN,and as a basis for comparison. Based on the input data (harvest date, initial moisture content, air flow, height/diameter ratio of the silo, presence or absence of heating, and meteorological conditions) the model predicted the percentage ofgrain deterioration, the energy consumed in ventilation and heating, the endof- drying date, and the amounts of grain with moisture contents over 20% or over 18%. Nine trials were run with different numbers of processing elements in each hidden layer, learning rules (delta-rule and normalized cumulative delta-rule), transfer functions in the processing elements (sigmoid or hyperbolic tangent), learning cycles (100,000 to 500,000), and data for learning. The ANN showing the best performance was the one with 20 processing elements in the first hidden layer and lOin the second, 100,000 learning cycles, hyperbolic tangent transfer functions and the normalized cumulative delta learning rule. Overall, this ANN satisfactorily reproduced the results of the procedural model. Generally, the average relative deviation was between I%and 3%. There were, however, a few extreme cases where this deviation was close to 50% for the deterioration. For the other variables, the maximum relative error was between 20% and 25%. If the ANN is to be used for decision support, these extreme values will have to be taken into account.
Keywords:
Résumé:
Mots-clés:
Citation: R. Lacroix, R. Kok et S. Fortin 1999. DETERMINATION DE L'APPLICABILITE DES RESEAUX DE NEURONES POUR SIMULER LE SECHAGE DU MAis EN SYSTEME MIXTE. Canadian Agricultural Engineering 41(2):105-112.
Volume: 41
Issue: 2
Pages -
Contributor:
Date: 1999
Technical field:
Conference name:
Session name:
Other information:
Type: Text.Article
Format: PDF
Publication type:
Source:
Relation:
Coverage: Canada
Language 1: en
Language 2:
Rights: Canadian Society for Bioengineering
Notes:
Other files:
©
2024
CSBE-SCGAB