Conclusions
Statistical instruments based on artificial neural networks were
developed and efficiently trained to be used both for the prediction
of wine antioxidant activity and for revealing the wine variety of
grapes, harvest year and originating winery.
The ANN having single hidden layer architecture of feed-forward
type was designed and trained, using the back-propagation
algorithm, with the aim of predicting AO activity. As shown by
the ANN simulation results, the prediction quality was good for
the testing subset of data. This capability of the ANN may spare
measuring costs and provide a useful model for specific wine processing
equipment design and optimisation.
Probabilistic ANNs have proven good classification ability. We
were able to demonstrate the complex but hidden relationships
between the concentration of total phenolic, flavonoids, anthocyanins
and tannins, associated with either inferred or directly measured
antioxidant activity, and wine variety, harvesting year and/
or winery. The results performed on the sets of data produced no
errors in identification. Thus, these ANNs proved to be reliable
software tools for assessment or validation of the wine essential
characteristics and authenticity.
Results of this research may be further used to establish a database
of analytical characteristics of commercial Romanian red
wines from various varieties, production year and wineries.
Conclusions
Statistical instruments based on artificial neural networks were
developed and efficiently trained to be used both for the prediction
of wine antioxidant activity and for revealing the wine variety of
grapes, harvest year and originating winery.
The ANN having single hidden layer architecture of feed-forward
type was designed and trained, using the back-propagation
algorithm, with the aim of predicting AO activity. As shown by
the ANN simulation results, the prediction quality was good for
the testing subset of data. This capability of the ANN may spare
measuring costs and provide a useful model for specific wine processing
equipment design and optimisation.
Probabilistic ANNs have proven good classification ability. We
were able to demonstrate the complex but hidden relationships
between the concentration of total phenolic, flavonoids, anthocyanins
and tannins, associated with either inferred or directly measured
antioxidant activity, and wine variety, harvesting year and/
or winery. The results performed on the sets of data produced no
errors in identification. Thus, these ANNs proved to be reliable
software tools for assessment or validation of the wine essential
characteristics and authenticity.
Results of this research may be further used to establish a database
of analytical characteristics of commercial Romanian red
wines from various varieties, production year and wineries.
การแปล กรุณารอสักครู่..