• Locked Locked
  • Replies 1 reply
  • Subscribers 82 subscribers
  • Views 1704 views
  • Users 0 members are here
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Calibration, Validation Data Collection Standards (WaterGEMS)

Jeremy Nichols
Offline

Hello all, 

In a recent discussion a few points came up regarding ensuring accuracy and limiting variables in SCADA and field data to be used in calibration & validation of a model. 

For our past clients, we have ensured that days chosen for calibration with an EPS model should be dry, high-usage weekdays, with no additional uncountable flow (flushing, etc). 

In a recent case, rain was prevalent throughout the data collection period and rainy days were used for both calibration and validation cases. The argument against using rainy days for system calibration is that irrigation usage may be suppressed, thus deviating from the target 'Average Day' case. Does anyone have ideas regarding additional effects that a rainy day may have on system behavior and customer usage? Would these effects be great enough to invalidate the test case? Obviously it is a very different story for runoff, sewer, and surface water models, but this particular example is a pressure pipe distribution system. 

Are there any other variables, environmental or otherwise, that are considered critical to control for when obtaining good calibration data?

Thanks, 
Jeremy 

  • Offline
    Jeremy,

    Your question is a very good one.

    If you calibrate an EPS model for a given day, then it is most valid for that day. As water use patterns deviate from the day, the model becomes less valid.

    The bad news is that no two days are identical. Who knows who is flushing their toilet or taking a shower at 10 am? I did an analysis of demand patterns once and found that there can be up to about a 15% difference in demand at any time of day for no real reason—just random human behavior.

    The good news is that these variations don’t impact the model results very much. Especially, in North American where fire flow requirements are large, the velocity and hence head loss in most pipes are small so that variations in daily demand have only a small impact on pressure. The biggest impact you’ll see is that tanks will fill slower and drain quicker as the demand increases. This shows up in the model as time shifts in pump on-off times. But the results in terms of hydraulic grade, pressure, etc. are still quite good.

    It would be great if you could calibrate the model over several significantly different kinds of days. This is more costly but as you move past the first day, latter days are not nearly as difficult to calibrate.

    Model calibration is not a black and white situation. No model is perfect. As the saying goes, “All models are wrong but many are still useful.” Models fall on a spectrum from perfect to awful. The better job you do with calibration and the more the condition you are analyzing approaches that for the calibration day, the closer you are to perfect.

    The best approach is to run the model as a real-time model. Every day, you get the previous day’s SCADA data and compare it with the model. Some days will agree well. On days that don’t, you can identify the source of the difference. Was there a fire, an incorrectly closed valve, a pipe break, flushing, unusual weather, special events, a new pump, new land developments, a large shutdown, bad SCADA signal, etc. As you do this, your confidence in the model and SCADA system increases and operators begin to understand and rely on the model. Then, when there is an operational problem the model is ready to go and support decision making. Our SCADAConnect tools are great for this. The more you use the model, the better it gets.

    Tom
Communities
Support and Services
Training and Learning
Social Media