The Right Demand Profile for Operational Decisions
Water utilities lean on hydraulic models at the moments that matter most.
Sometimes it's planned.
A pipeline is coming out of service for maintenance, and operators need to know the impact before they close a valve.
Sometimes it isn't.
A main breaks, and operators need to know quickly which customers are affected, how storage will respond, and what they can do about it.
In both cases, operators turn to a planning engineer with hydraulic modeling expertise.
The model’s insights shape early decisions in terms of which customers will be impacted.
It's a role where precision matters, and where a model’s output is only as good as its input – the demand profile.
Which raises the question the modeler faces - what demand profile should the model use?
The standard profiles
Most Australian hydraulic models run on a small set of standard demand profiles - average day, 95th percentile, or peak day.
These are reasonable, well-understood, and for many planning scenarios they're adequate.
But real events rarely land neatly on one of them.
Consider a large pipe failure on a warm 32°C (90°F) day. It's not an extreme heatwave, but demand is above average.
The modeler faces a choice:
use an average day profile and risk understating the impact
use a peak or 95th percentile profile and risk overstating it, or
modify an average or peak day profile with a spreadsheet under time pressure.
And what if the scenario falls on a public holiday, when demand profiles shift again?
None of these are wrong, exactly.
But each involves reaching for the nearest standard profile and accepting that it doesn't quite match the reality of the demand conditions.
SensorClean’s profile builder
SensorClean produces profiles from subsystem data which offer more localized water behavior compared with profiles developed from source data (see blog about variability across water networks).
The modeler selects the day of the week – including public holidays – and the forecasted temperature.
These inputs go into regression analysis – which leverages the subsystem’s historical relationship between cleaned consumption data, temperature, antecedent rainfall (dryness) and days of the week.
The modeler then gets a profile which is much more representative of what’s actually going on in the subsystem where the planned/unplanned work will occur.
The profile can also reflect the subsystem's actual leakage, measured from its night flow, rather than a flat unaccounted-for-water allowance applied across every system.
Night flows change all the time and adopting fixed leakage percentages isn’t representative of reality.
Going finer with regression
Where the data supports it, we can do even better.
Using another regression model, demand can be separated for houses and apartments (separating houses and apartments blog).
Smart-meter data helps here too. Even where the record is too short to capture many Maximum Day Demand events, it's still a valuable line of evidence for characterizing everyday demand behavior.
The result is a demand profile grounded in the network's real composition, not a one-size-fits-all profile.
Knowing what drives high demand
Data mining also reveals when a subsystem is most exposed - and why.
For example, in one subsystem, the top ten Maximum Day Demand events all fell on weekends and public holidays.
Seven were driven by hot, dry conditions; the rest weren't.
That's useful for an operator to know before scheduling work - a planned shutdown on a weekend in that subsystem may not be the best idea.
Knowing the drivers of high-demand events - and the days they tend to land on — turns demand data into a scheduling decision, not just a modeling input.
Stress-testing with real failure data
SensorClean also quantifies the water lost during large failure events, and that has a use here too.
During a planned shutdown scenario, the modeler can go a step beyond the base case.
Using a real, measured failure volume from the network's own history, they can simulate what happens if a main fails while the system is already down for maintenance.
It's a low-probability scenario, but a high-consequence one - and it's cheap to test.
Worth knowing before a valve is closed.
More evidence, less risk for operators
A demand profile matched to the conditions gives operators faster, better-grounded answers — whether the event was scheduled for the week ahead or is unfolding in the moment.
Better inputs, better decisions, less risk to the utility and its customers.
Average day demand is a fine place to start. But when the best evidence is already in the data, why not use it?
If you'd like to see the profiles your own subsystem data can produce, get in touch.