3.1 Definition of used performance ratios
Some statistical data were established as comparison criteria between the methods for estimating parameters.
In this work, parameters from the retention of coefficient Ki(T) were optimized, considering two different performance rates for an arbitrary i component:
(1) the sum of the square error (sSE), Eq.
(2) the maximum error in absolute value (MEAbs) Eq, :
MEAbs =max top)
The sum of the square error (SSE) is directly related to the coefficient of determination R2, wherein the minimization of the former implies the maximization of the latter. R2 is a standard approach measure, contrary to what happens with the SSE, which is discrepancy between the measured data and the estimated model. However, the analyzes described in this paper are shown in terms of R2 values, even though it was the SSE that which was the minimized criterion. R2 values of the order of 99.9 or 99.99 usually indicate a good approximation.
In the case of the maximum error in absolute value (MEAbs), instead of see or R2, a well-defined meaning can be found, because it is known from the value of MEAbs, the worst case of all temperature settings for a given component.
Considering that retention times obtained on Part l of this series were the order of minutes, prediction errors of tr around a few seconds are considered acceptable.
Even though the highest relative error [%] in MEP module has not been directly optimized in parameter estimation, it was also calculated according to Eq . MEP values lower than 1% are described in literature as indicators for good prediction performance.
สูตร
It is noteworthy that for the evaluation of any one of the objective functions described, the model for retention time prediction must be solved NExp times.