despite several steep changes in the ethanol profile, closely
matched predicted and real concentrations (Fig. 1). Both
FFANN and RANN estimators were unbiased, as their
mean errors were smaller than the reference method measurement
error and returned a random distribution around
zero (not shown here). In summary, the best soft-sensor
configuration for lab distillations provided reliable predictions
with accuracy as good as 0.6% v/v (sum of reference
measurement error and MAE).