Page 407 - Design of Simple and Robust Process Plants

394 Chapter 9 Operation Optimization
9.6.10
Implement Data Reconciliation: Step 10

Data reconciliation is applied for:

. Performance meter (step 2)
. Estimation of input data for the simulation such as feed flow and composi-
tion, energy flows
. Gross error detection
All these functions need to be tested and the criteria set before moving to the next
step.

9.6.11
Implement Simultaneous Data Reconciliation and Parameter Estimation (DR and PE):
Step 11

DR and PE is performed to achieve a close fit between simulation and the actual
operating process. The reconciliation and estimation is only possible in case of
redundancy in measurement. Often, redundancy is insufficiently available, and in
that case the parameters are calculated values based on process measurements. The
technique of DR&PE is identical, and is described in Section 9.5.2. The parameters
might be capacity-dependent. The procedure followed is that the parameter is deter-
mined at the current state of the process and updated every optimization cycle. This
includes an off-set in the parameter if the capacity of the process has changed sig-
nificantly, although at the next optimization cycle the parameters are updated. The
selection of criteria for gross errors of measurements are based on the same criteria
as for the data reconciliation module. Decider-3 also has criteria for outliers between
simulated and actual measurements. The latter can only be selected after the model
validation (step 12)has been performed.

9.6.12
Validate Model: Step 12

Model validation is an activity designed to minimize the difference between simula-
tion and actual performance (J. Krist et al., 1994). The assumption for the validation
is that the unit models were initially verified and the model has been demonstrated
to be robust. The validation is performed in different steps:
. Gross modeling error detection.
. Smoothing of the model, by data reconciliation (DR)and parameters estima-
tion (PE)on an extended set of measurements.
. Overall model validation, by comparison of simulated performance with mea-
sured performance.