Weights ¬ |
||
---|---|---|
In the Weights Node, the measurements to be reconciled in the Estimate run are selected. This node also provides the means to automatically generate weighting factors based on the selected error assumption.
Tabbed ViewsActions Tabbed Views Weighted Measurements Advanced History
Weighted Measurements
Top of Topic This view displays the measured items that are available for reconciliation and options to weighting them, as explained below for each available grids:
Grid Views Reconciled Measurements Weight Generation Options
Reconciled Measurements
Top of Topic When a variable is selected for reconciliation in this grid, REX will attempt to match that measurement by optimizing the kinetic parameters. When a variable is unselected, REX does not make an attempt to match this measurement, however the weighting factors are still preserved. This allows you to conduct estimation studies with different sets of reconciled measurements, without having to change the weighting factors. When Pressure is defined as Free in Reactor node for a reactor with one gas phase, you may reconcile the pressure measurements by selecting Pressure as a Reconciled Measurement in this node. Thus, the kinetic parameters are adjusted to match the pressure measurement. When Pressure is selected as Controlled in the Reactor node, REX calculates the value of the floated flow to match the pressure, while the other selected variables are reconciled by adjusting the kinetic parameters. Temperature may be reconciled only if Use Energy Balance is selected for calculating temperatures in the Reactor node. Note: A given compound can be selected to be reconciled only if that compound was included in the Measurements node. The same holds for pressure, temperature (when enery balance is enabled), pseudo-compounds and derived quantities.
Weight Generation Options
Top of Topic This grid is used to generate the weighting factors based on the nature of the error assumption selected. After selecting the desired error assumption option, you may generate the weighting factors by pressing the AutoGenerate button. If you choose to enter the weights manually, select the reconciled items in the Reconciled Measurements grid and then directly enter the weighting factors in the Weights → Sets node. For all the options of error assumptions, if the Ignore Zero Values is selected, then the weighting factors are set to zero for those measurements whose values are specified as zero. These measurement points are also excluded from the unweighted deviation (LSQ Errors) as indicated in the Results: LSQ Error section. If the Ignore Zero Values option is unchecked, the weighting factors for these zero value measurements are set to the highest value among the nonzero measurements for the corresponding compound and experimental data set. In this grid, if you select "Generate Weights for → All Included Data", the weights are generated for:
In some cases, you may have specified weights for some experiments, and then included new experiments or changed the experimental values for some sets. In order to generate weights only for the new or modified data without affecting the weights that were previously loaded, you may select the option "Generate Weights for → Only Modified / New Data". If you wish to check which sets are new/modified, please go to the Advanced tab and use the Select Modified Data Button. After setting the above options, please click on the Autogenerate button to generate the weights. Further details on the weighting strategies are provided below. Explanation of Error Assumptions. The kinetic parameters are estimated by the Weighted Least Squares method, where the objective function to be minimized is as follows: Objective = w1*(Exp1 - Cal1)2 + w2*(Exp2 - Cal2)2 + ... wn*(Expn - Caln)2 where wi are the Weighting Factors for the measurements; Expi are the Experimental Data Values, and Cali are the Calculated Values at the Data Points. REX provides the following options for the error assumptions. Select the appropriate option to generate the weighting factors. You may always fine tune them later.
This means that all measurements have same accuracy (or same absolute error). Therefore, on Autogenerate with this option, all the data points are equally weighted with a value of 1. In this case, the results from the REX estimation may show model deviations (difference between measured and calculated values), which are similar across a large range of measurement values. As an example, lets consider two data points, with measurement values of 0.01 and 1 units, and calculated REX values of 0.02 and 1.01. In this case, both points result in a mismatch of 0.01 units. Since the weighting factors (wi) are all 1, both of these deviations are considered to be identical by the REX estimation program. This means that this weighting method is not suitable if you are concerned about the high percentage error (100%) for the first data point. In general, with this method, large values tend to get a better match on a relative (percentage) basis, whereas smaller values may be poorly matched on a relative basis. This means that the measurements have uniform percentage errors. For the example shown above, the weighting factor would be 10000 and 1 for the first and second data points respectively. This means that REX will place 10000 times more importance on the square of model deviations for the first point versus the second point. Thus, selecting this option will usually result in smaller measurements getting significantly higher weighting factors, since REX attempts to ensure that the model predictions preserve the uniform percentage error assumptions. This may not be desirable if the small measurements are considered unreliable, which may be true for some chemical systems. While the uniform absolute error assumption tends to overweight larger measurements, and the uniform percentage error assumption tends to overweight smaller measurements, the hybrid option offers an intermediate alternative to the above two options. This is the recommended and most commonly used option. Advanced The view is used to modify the weighting factors for a collection of variables and sets in one action.
Grid Views Select Sets and Variables Advanced
Select Sets and Variables
Top of Topic Here, you may select the variables and sets for which the weights will be generated. The following columns are available:
Advanced
The actions grid allows to specify the collection of items to be chosen for modification. The options available are:
Depending on the selections above, more information is requested by REX as follows:
History
Top of Topic This view shows the weights modification history, thus documenting the weighting strategies. Since the estimated parameters are often sensitive to the selection of the weighting factors, this history is useful to communicate the logic behind the selection of the weighting factors. You may also add your notes here for documentation purposes. Actions Quick Run Open Solver Top of Topic See Also: |