Uncertainty and gross error detection

I have a question about data uncertainty and gross error detection.

Please find my model attached, 150 flows and about 90 processes.
At the moment I only entered final data for subsystem "P01 Waterbehandeling" (=water treatment).

The flow data (in t/d) is based on data from flow meters over a 2 year period.
From this data I calculated the mean, the standard deviation and number of data points (typically daily). I then calculated the standard deviation of the mean (stdev of sample / sqrt(number data points). I used this as the standard uncertainty in the data model.

I'm getting gross errors on most of the flows from the subsystem (P01).

How do interpret these gross errors?
Can I get more detailed information for every flow about how much he's out of the expected uncertainty range (I see there are z-values in the detailed trace, but how do I interpret these)?
What is the best way to relax this model?
Do I relax the uncertainty for every gross error flow, or only those flows with the largest z-value?
Can I relax the model overall (general alfa value for probability that value is gross error)?

Best regards,

Geert

Dear Geert,
the z values you mentioned are not related to a statistical test. They refer to a column of the coefficient matrix (needed for the nonlinear data reconcilaition algorithm) that contains the residuals of the (transformed) model constraints.

If you want to display something similar to the z-values, you can use an undocumented feature of STAN (implemented for testing reasons):
(1) Right-click the STAN icon on your desktop and select properties, then tab shortcut.
(2) Add "/debug" at the end of the shown target:
"C:\Program Files (x86)\inka software\STAN 2.6.801\Stan.exe" /debug
(3) Click OK.
(4) Start STAN and press ALT+F5 to open the Calculation window.
(5) Enter a desired k-factor (= absoute standard scores = abs(reconciled value - original value)/standard uncertainty of original value).
(6) After calculation, in the detailed trace output (click button "show text messages"), those flows will be displayed where the k-factor is larger than the stated k-factor.

The discrepancies between input and output flows of your processes are too large compared to their assigned uncertainties. I used the feature "Display coarse balances of processes" ("Extras > Options > Tab Processes & Flows > Region Processes") to highlight the problem for process P11:

There are 77.76 t/d more going in than coming out but you stated the input and output values to be pretty precise. So, either your measurements are of poor quality or the assigned uncertainties are to small.

Hope this helps to solve your problem.
All the best,
Oliver

Hello Oliver,

thank you for this idea about the k-factors.
I export the data in excel and calculate the k-factor there, makes it easier to filter and compare.


Best regards,

Geert