Does traditional calibration hide errors in your demand analysis?


Courtesy of Innovyze

Untitled Document

A well-calibrated water distribution model can address numerous problems encountered in the everyday operations of a water utility. However, without sound calibration and validation, any model remains unproven. The calibration process is fundamental to the accuracy and usefulness of the model.

Calibration is essentially the process of explaining any differences between model results and the observed values. The traditional approach is to address these differences by adjusting the roughness factors of pipes. As pipes age the roughness of the internal surfaces increases and flow is impacted. Engineers use tables of standard roughness factors depending on the age, diameter, and material of pipes, and these values can be used as the initial estimate of the roughness. If model results and telemetry readings still do not coincide, it is accepted practice to keep changing the roughness values until the differences between the two sets are reduced to an acceptable level.

However, adjustment of roughness should not be used, as it sometimes is, to account for all the errors that may be present in the first build of a water distribution model. Although wrong roughness factors are often the reason for the differences in observed versus predicted pressures, in many models errors in the assumed demand /flows have a far greater impact on results than errors in roughness estimates. A careful analysis of demand is too often overlooked as an essential activity before the roughness calibration begins, an analysis that can be conducted area by area, with the objective of fully understanding all the components of demand and their differing diurnal profiles.

Demand Area Analysis

InfoWorks WS provides an automatic tool, Demand Area Analysis, to address the issues of accurately estimating demand at local level, area by area. The approach involves looking at the flow balance of each area - the amount of water that flows into the area and does not flow out – measured using flow meters.

The total volume of the flow balance can be divided into three components:

  • Known demand – metered demand, which usually covers commercial and industrial demand plus in some cases some or all domestic demand
  • Un-profiled demand – un-metered demand, plus any estimates of illegal connections which are assumed to follow the same profile as domestic demand
  • Leakage – leakage as a percentage of total supply is almost always in double figures, in some systems accounts for 40-50% of the total production, and in extreme cases up to 80%. In most cases leakage is by far the greatest contributor to unaccounted for water (UAW) unless illegal connections are a big problem.

The key to identifying the profiles and volumes of each of these components of consumed demand in an area is to start with a 24-hour plot of the flow balance – the total volume of water. Metered demand – known demand - can then be attributed accurately under the curve, hour by hour. The next step is to estimate leakage. The recommended approach is to take the point during the night, perhaps 3am, when consumption is at its lowest, and make the assumption that un-metered domestic demand is a fixed, low amount - usually 1.7 liters per property per hour. Now that known demand is allocated and un-profiled demand assumed, the third component, leakage, can be estimated accurately as the remaining component of the profile.

The 24 hour profile of leakage is then extrapolated from the single minimum night consumption point level by using a pressure/leakage relationship, leakage being pressure related. InfoWorks provides two alternative functions relating pressure and leakage. With a good understanding of the level of leakage, including its diurnal variation, un-profiled (un-metered) demand can be estimated, and the analysis of the three components of demand is complete.

Once both leakage and un-metered consumption volumes and daily variation are identified, they can be applied to network elements. Typically, leakage is attributed to selected pipes within the area proportional to pipe lengths. However, in systems where leakage occurs on different parts of the system, such as leaky house connections, leakage can be applied to nodes, potentially proportional to the number of house connections. Similarly, un-metered water is assigned to selected nodes, typically proportional to the number of house connections. Both leakage and un-metered consumption are assigned daily variation patterns identified in the demand calibration. Using this process a zero-balance in flow is achieved for each area and the model can be run and the process of pressure calibration started.

Calibration and Roughness Factors

The first stage of the calibration of the model is to input the roughness values from the tables, based on age, material, diameter, and condition of pipes – a process InfoWorks carries out automatically. A run of the model may still identify differences, but if large differences are observed these are usually the result of false assumptions in the network model - for example valves assumed open may be partially or totally closed. These errors should not be handled by changing roughness factors but by changing the underlying model.

When the network model has been checked and anomalies explained and corrected, the final calibration can be completed. Not all pipes will have the book values of roughness as attributed by the tables, and the second calibration can address this to produce an accurate and proven model.

The role of InfoWorks

Good modeling software such as InfoWorks WS can assist with all the above processes.

Demand Area Analysis, supported within InfoWorks WS, is a process that hitherto required spreadsheets. The results of the analysis, the profiles of profiled demand, un-profiled demand and leakage, can be input into InfoWorks WS as part of the demand input. Other aids to building demand inputs to the model are the import of Address Points, which are then automatically allocated to the nearest logical pipe. Property based demands with related consumption, average factored demands, and direct profiled demands can assist in the demand analysis, enabling swift predictive analysis of leakage.

Good calibration required a software suite big enough to run an all mains and all valves model is very useful – if skeletonization is being used together with calibration, the model becomes several removes away from the real world. InfoWorks WS handles very large models easily.

Another InfoWorks tool that helps with calibration is the Auto Calibration module. This inputs the standard roughness figures based on tables held in the software, and then adjusts the factors to arrive at a minimum sum of differences (model versus SCADA data) in pressure observation points. In average models (a few thousand nodes) this method outperforms the various ´blind´ numerical methods that randomly change friction factors seeking a good fit. In large models (tens of thousands of nodes) with many degrees of freedom within which to optimize, this method can be the only sensible approach.

In summary – calibration is an essential part of modeling, and the best software has tools to help in this process. But calibration should begin only after the demand model is fully understood, and the model cleaned for major errors. Once these stages of model building have been properly completed, the standard roughness factors from tables may be all that is required to calibrate the model.

Customer comments

No comments were found for Does traditional calibration hide errors in your demand analysis?. Be the first to comment!