ALCOHOLIC FERMENTATION UNDER ENOLOGICAL CONDITIONS - USE OF A COMBINATION OF DATA-ANALYSIS AND NEURAL NETWORKS TO PREDICT SLUGGISH AND STUCK FERMENTATIONS

The possibility of predicting sluggish fermentations under oenological conditions was investigated by studying 117 musts of different French grape varieties using an automatic device for fermentation monitoring . The objective was to detect sluggish or stuck fermentations at the h alfway point of fermentation. Seventy nine percent of fermentations we re correctly predicted by combining data analysis and neural networks. data by using precise kinetic data (instantaneous CO, production rate ) and by testing a large number of musts. Our objective was to predict the risk before or at the halfway point of fermentation, when nutrien t additions are still efficient. We first tested a combination of data analysis and linear modelling. Then we replaced linear models by neur al networks in order to improve precision.