Product(s):	WaterGEMS, WaterCAD
Version(s):	08.11.XX.XX and higher
Area:	Modeling
Original Author	Dr. Tom Walski, Bentley Systems

Introduction

The purpose of this article is to provide guidelines and best practices for hydraulic model calibration of water systems.

Model Calibration and Darwin Calibrator

When a water distribution system model is first constructed, the model results need to be compared with field data to ensure that the model accurately represents the real world. Initially, most models do not agree well with the field data. The model inputs must be adjusted (or in some cases incorrect field data must be discarded or corrected) to bring the model into acceptable calibration. Calibration is defined in attachment #1 at the bottom of this article.

The difficulty with calibration is that there are many reasons that a model and field data will not be in agreement. The primary problem is not making the adjustments, but understanding what is causing the discrepancy in the first place. Below is a list of some (but not all) of the things that can make the model and field data differ. You need to be a bit of a detective to determine which applies in your situation.

Physical
- Pipe size/location
- Pipe connectivity
- Pipe roughness*
- Pressure zone boundary
- Pump curves
- Pipe material/age in GIS
- System changes since model built
- Elevation data
- Over-skeletonized model^#
Operational
- Valve open/closed/throttled status
- Control valve operation/settings
- Transient events
- Actual operations not matching control rules
- Unusual operations when data were collected
- Tank water levels
- Pump status/speed
- Lack of sufficient sensors/gages
- Water quality reaction rates
Demands
- Spatial allocation
- Model does not reflect conditions when data collected
- Large customers with atypical demand patterns
- Not accounting for seasonal changes in demand
Data
- Inaccurate/uncalibrated gages/meters
- “Latched” data from SCADA
- Understanding SCADA data – average vs. instantaneous

*Note: typically roughness does not have a dramatic effect on carrying capacity with few exceptions such as cast iron. As an example, the City of Charleston, SC, which has the oldest cement mortar lined pipe (dating back to 1922), conducted a C-factor test (see YouTube video at the bottom of this article). The test showed that the roughness had not changed over 97 years. So, if Darwin Calibrator recommends a C-factor of 80 for a cement mortar lined pipe, it is probably compensating for something else such as a closed pipe or mistake in data entry somewhere.

^#Note: when performing calibration it is best to have a model that is as close to the real system as possible. For this reason having an over skeletonized or simplified model would lead to unaccountability of the entire carrying capacity of your system. Using a model that accurately represents the system with all its pipes would help in getting better calibration results.

It takes considerable judgment and experience to identify the source of the discrepancies. Users are often overwhelmed by the possible choices. It is best to take a logical set of steps through the calibration process. Attachment #2 at the bottom of this article provides a procedure to approach calibration.

Darwin Calibrator (available in WaterGEMS) provides a tool to adjust the first three parameters from the above list once the user has identified that one of them is the source of the discrepancy. If a user decides to adjust the wrong parameter in order to make the model look calibrated, it is referred to as “Calibration by compensating error” which uses one error to correct for another. It may be acceptable to assume that a certain parameter is the sole source of discrepancy and allow Darwin Calibrator to make adjustments, but these adjustments must be used with caution.

Darwin Calibrator uses the head loss between the water source as the primary driving force for its solution. It is therefore essential to ensure that the head loss between the source and the pressure measuring point is significantly greater than the error in measurement. For example, if the head loss from a tank to a pressure gage is 1 m, and the error in measurement is +/- 2 m, this data should not be used. For more on data accuracy, see attachment #3 at the bottom of this article.

Before beginning to work on calibration, users are encouraged to read the attachments to this article (see "Attachments" section further below) and complete the Bentley training classes on Calibration and Darwin Calibrator:

(Bentley LEARN Server) Water Distribution Design and Modeling Advanced using WaterGEMS CONNECT Edition (see "Automated Calibration" on-demand lecture and workshop)

(OpenFlows YouTube channel) Water Advanced Training (WaterGEMS) (See part 1 and 2 on Darwin Calibrator)

Tips for Darwin Calibrator

Once you are ready to use Darwin Calibrator in WaterGEMS, consider these tips:

Keep in mind that all pipes within a particular roughness group will be adjusted together, not individually. It is up to the engineer how many groups to create. The recommended workflow is to group elements that have similar characteristics such as age or material, since it can be assumed that two pipes of the same age or material will see similar changes to the actual roughness (as again, all elements within the groups are adjusted together, not individually).
Read "GA-Optimized Calibration Tips" and "Darwin Calibrator Troubleshooting Tips" in the WaterGEMS Help.
For performance-improving tips, see: Darwin Calibrator Performance Improvement Tips for Large Models
After computing a calibration study, you will sometimes see results that do not have a good fitness or do not make sense. Below are a few general tips to look at. More information on Darwin Calibrator can be found in the WaterGEMS Help documentation.
- Try running a manual calibration study with no adjustments to the roughness or demand. In other words, keep the multipliers for the adjustments groups at 1. Then look at the fitness. If you have a very high fitness number, this is an indication that something is wrong with the model setup. Review your model and make sure the results look reasonable and that no unusual user notifications are generated.
- Verify that the boundary overrides and demand adjustments are properly entered. This is very important since otherwise Darwin Calibrator will use values in the model at the times of the field data snapshots instead. Since the results in the model results might not be the same as what is actually occurring in the field, it is important to make these changes. Accurate data model-wide is important as well. Good, accurate data will mean the calibration study will work optimally.
Using field data that includes a higher flow tends to allow you to find better results for the roughness coefficients. With higher flows, you have higher velocity results. Because of the structure of the Hazen-Williams equation, this will allow you to have a larger spread in C values for a difference in pressure.
If you are computing an optimized run, you can change some of settings in the Options tab. Highlight the optimized run on the left side of the window, then choose the Options tab on the right. Detailed information for the options can be found in WaterGEMS Help. Changing the Maximum Trials and the Non-Improvement Generations items to higher values will allow Darwin Calibrator to try more adjustments and possibly find better solutions. You can also change the Fitness Tolerance and the Solutions to Keep. It is recommended that the Advanced Options remain set to the default values.
Since Darwin Calibrator’s genetic algorithms can only compare a result to results that came before, making changes to the options can create new sets of solutions. It is recommended that you run several optimized runs even if the initial calibration study is good. A better solution might be generated after changing the parameters.
WaterGEMS and WaterCAD also come with several sample models that have completed manual and/or optimized calibration studies. These are a good resources to see the general setup of a calibration study, and can allow you to calculated models and see how the results differ not only with between manual and optimized runs, but also in changing the options. The sample files can be found in the "Samples" subfolder within the installation folder.
Example5.wtg is an example of calibration based on three hydrant flow tests done at different times of the day. The hydrant flow, pressure, and corresponding pump and tank status are shown in the Demand Adjustments, Observed Target, and Boundary Overrides tabs within the calibration study, respectively.
Example4.wtg shows a system separated into DMAs, demonstrating the use of the Pump Station element. A Darwin Calibrator study is included, giving examples of calibration based on both static and residual hydrant flow tests. Optimized runs where both demand and roughness can be adjusted are included, along with examples of Manual runs where roughness and/or demands are adjusted manually.
Example8.wtg is an example of a leakage detection study using Darwin Calibrator. Observed hydraulic grades at several different locations in the system have been entered, for several different times of the day. The "Detect leakage node" option is used in each of the optimized runs, to detect leakage node candidates by using emitters.

Attachments:

Attachment #1 - (JAWWA, 2013) - Defining Calibration

Attachment #2 - (JAWWA, 2017) - Procedure for Hydraulic Model Calibration

Attachment #3 - (JAWWA, 2000) - Model Calibration Data - the Good the Bad and the Useless

Water Model Calibration Tips

Introduction

Model Calibration and Darwin Calibrator

Tips for Darwin Calibrator

Attachments:

See Also