USING EXTENDED PERIOD SIMULATION MODELS

FOR DESIGN AND OPERATION

Thomas M. Walski

Abstract

Extended period simulation (EPS) models have been somewhat unpopular because the
huge amount of data they produced made it difficult to understand the results of an EPS
run. With the coming of improved graphical output, use of EPS models is much more
practical. Some practical examples of what to look for in an EPS run are presented.

What is an EPS Model?

An extended period simulation model determines flows and heads in water

distribution system over time. This is done by linking steady state solutions by keeping
track of tank water levels, pump operation and fluctuations in demand. An EPS run can be
viewed a as movie of what is occurring in the water distribution system.

Why Use EPS Models?

Steady state water distribution models are the most commonly used tool in the
engineer’s arsenal for system design. They can answer a wide variety of questions on pipe
and pump sizing and system layout. They require less data than an EPS model. However,
there are some studies where steady state models are of limited use and extended period
simulation (EPS) models are required. Some examples of these problems include:

1. checking tank volumes,

2. evaluating pump cycling for tank filling and draining,
3. analyzing energy usage,
4. preparing for special operations such as emergencies or shutdowns,
5. operator training..

This paper will discuss what to look for in reviewing EPS output.

In using steady state modeling, the engineer tends to focus on the worst case
situations (peak hour demand, catastrophic fire) since these usually control design
decisions. However, water system operation is not truly steady and in looking only at
worst case situations the engineer may miss some other problems (e.g. tank refilling,
overpressurizing some areas). EPS runs provide a way to identify such problems.
EPS Setup

The key decisions in setting up an EPS are how long to run and how fine to divide
up the time steps.

Duration. The duration of the run needs to be long enough to capture the events of
interest. Many modelers use 24 hour durations on the supposition that subsequent days
will mirror the 24 hour period. However, this is not always true. The initial conditions
(i.e. tank levels, pump operation) tend to distort the results somewhat and only by running
at least two days can a model user be sure that subsequent days will be like the first day.
Some problems (especially those dealing with tank refilling) may not show up until the
second or third day. Some tanks may take several days to recover after a fire while some
small tanks (e.g. hydropneumatic) may go through a large number of cycles with an hour.

Time Step Size. The best time step depends on the natural fluctuations in the distribution
system and are site specific for each system. If pumps are cycling every 20 minutes then
time step no longer than 10 minutes is needed. If the pumps are cycling roughly every
three hours, then an hour time step is acceptable. If the system runs steadily with only
minor fluctuations in demand, it may be possible to use 3 hour time steps.
Long durations with small time steps and large systems without much
skeletonization tend to result in somewhat long run times. It is important that the model
user have adequate computer power to run the model effectively. For the right
combination of duration and step size, the user needs to consider the nature of the
problem being addressed.

What to Look for in Output

Because they calculate information for every node, pipe, pump , tank and valve for
each time step, EPS models produces voluminous output. Computers are great for
crunching numbers. Reviewing these numbers in tabular form for a real system, whether
by scrolling screens or paging through printout, is a daunting if not impossible task.
Computer generated graphics are the only way to quickly get a handle on what a given run
means. This paper will demonstrate what to look for in computer graphics.
Early pipe networks models and no graphics and later ones had graphics that were
not extremely user friendly. Some model users would try to export data to other graphical
software. Battling with the output was a challenge to anyone running an EPS model. With
recent models, viewing EPS results has become a quick, easy step in the modeling
process. The ability of the model to plot the results of several runs (scenarios) on a single
graph is especially valuable in design comparisons.
The key to using graphical output is for the user to ask, “What would I measure in
the distribution system if I had unlimited resources?” Rather than installing such data
acquisition systems, well calibrated models can give the same results at almost no cost.
The next section shows model results and discusses what the user should look for in the
graphical output. The graphs in this paper were prepared with Cybernet 3.0.
Evaluating Pumping Capacity. Usually the best indicator of how a system is performing is
a graph of tank level vs. time. Such a graph can provide insights on whether the system
will work when placed in service. Figure 1 shows the fluctuations in tank water level for a
system with three alternative pumps for a peak day demand. The curve labeled “Good
Pump” shows how the tank level would fluctuate if the pump were matched well with the
demands. The pump runs steadily and tank levels stay in the 10 ft (3 m) range set by the
modeler.
The curve labeled “Large Pump” shows a pump with excess capacity such that it
cycles on and off a few times, filling up the tank in a few hours and then remaining off for
the rest of the day. Such a pump would be good if the source (e.g. well, treatment plant,
clearwell) could keep up with the pump discharge. The curve labeled “Small Pump”
shows what would happen with an undersized pump. It would only work by letting the
water level drop to a point where it’s discharge matched the demand. Because of the way
in which pump curves drop off steeply, a pump does not need to be undersized greatly to
cause a shortfall in supply.

Evaluating Piping Capacity. A similar plot of tank level vs. time for several different main
sizes feeding the tank is shown in Figure 2. A 10 in. (250 mm) pipe can keep tank levels
above the target of 290 ft (90 m). A 12 in. (300 mm) pipe only slightly improves capacity
because once the velocity becomes small increasing pipe size makes little difference.
However, decreasing the pipe size to 8 in. (200 mm) results in inadequate tank filling rates
and excessive draining rates.

Tank Volume. Figure 3 shows how tanks with different volumes show up in EPS models.
The 30 ft (7.5 m) diameter tank shows a reasonable diurnal fluctuation, the 20 ft (5 m)
tank shows more cycling of the pumps feeding the system, but may still be satisfactory.
However, the 10 ft (2.5 m) diameter tank shows much more cycling and the system cannot
keep up with peak demands during several hours (15, 39 and 64) each of the three days
run.

Tank Elevation. While much of the work in determining the overflow elevation for a tank
can be done with steady state models, EPS runs can also provide some insights. Figure 4
shows three tanks with overflow at 300 ft (95 m). In the case of the “Base” tank, the
water level fluctuates through an acceptable range. In the “tank low” situation, the tank is
at too low of an elevation for the system and pumps and it is difficult to get the tank level
to drop without lowering pressure too much. In the “tank high” case, it is difficult to keep
the tank water level in the right range. Raising and lowering the tank overflow level in
model runs in design is much easier than trying to tinker with pumps and their operation
once a tank is erected at the wrong overflow elevation.

Low Pressure. There are always a few locations in any pressure zone that have low
pressure problems. EPS model output can help the user distinguish between low
pressures due to: 1. elevations and 2. excessive head loss. In the first case the solution
may be modifying pumping, PRV adjustments or moving the customers to an adjacent
higher pressure zone. In the second, more piping capacity, opening a throttled valve or a
tank near the customers is needed.
Figure 5 shows that customers at a high elevation will have pressure graphs that
are consistently low. On the other hand, customers with low pressures due to inadequate
piping capacity will have graphs with low pressure only during peak demand periods.

PRV Pressures and Flows. The graph a downstream pressure at a properly operating
PRV (pressure reducing valve) is fairly dull, a straight line with constant pressure.
However, a plot of flow through the PRV can provide information on flow rates that is
generally not available to the utility (because flow is not measured through most PRV’s).
An even more interesting situation corresponds to a PRV that is a backup feed
(sometimes called a “sleeper” PRV) to an pressure zone. The model can calculate when it
is operation and produce a graph of pressure vs. time in Figure 6. Figure 6 should match
the results of a pressure chart on the downstream side of the PRV. When the PRV is
closed, the pressure is dictated by other sources. When the PRV is in operation, the
pressure is a constant at the PRV setting.

Pump Controls. Some water companies like to keep their tanks full such that whenever
the level drop as much as a foot from the top, they turn pups on. The impact of
establishing “pump on” and “pump off” setting at a tank can be viewed with the model. A
graph like Figure 3 show a typical water level fluctuations. Figure 7 shows head
fluctuations in a pump discharge in an attempt to keep the tank full. A larger span between
“pump off” and “pump off” settings will reduce the period of these fluctuations and have
the benefit of reducing wear on the pump.

Tank Off-line. Occasionally, utilities must take a tank off line for inspection or painting.
When this happens a constant speed pump discharging into a dead end system can produce
very high pressures during low demand periods or low pressures during high demand. It is
a simple matter for the model to predict those pressure as shown in Figure 8. (In this
example, the pressure would normally be fairly constant at about 70 psi (550 kPa) when
the tank is on line.) With this information, the water utility can make decisions whether
they need to relieve the pressure while the tank is off-line.

Fire Events. While a water utility can estimate its fire flows with fire flow tests, it cannot
check how the system responds for the duration of a several hour, 1000 gpm fire (60 L/s).
With an EPS model, the utility can examine how a tank water level would drop during a
fire and how it would refill after the fire. In Figure 9, the tank barely stays above its 260 ft
(85 m) bottom during a fire from hour 2 to 4. It takes until nearly hour 20 for the tank
level to recover. If the fire had gone on for 3 hours, instead of 2, the tank would have
drained. This also could have been simulated with the model.

Pipe Breaks. A pipe break is similar to a fire in that there is a large flow for several hours
while the break is isolated and the tank is counted on the provide the extra water during
that time. However, in the case of a fire the tank begin to refill immediately after the fire.
In the case of a break, there is an initial time (hours 2 to 4 in this example in Figure 10)
where the leakage is very high (500 gpm [30 L/s]) followed by 10 hours in this example
where flows are restricted in an important main while the break is repaired. Only after the
break is repaired and the pipes placed back in service will the tank recover (after hour 14).

Application

The author was recently faced with the problem of siting a water storage tank in a large
pressure zone with very wide fluctuations in pressure (30 psi, 230 kPa) both between
seasons and over the course of an individual day. The was no readily available solution for
the correct elevation to set the overflow for the new tank. In addition, the pressure zone
was fed by two sources roughly 15 miles (24 km) with significantly different heads and
steep slopes in the area limited the number of potential tank sites. If the tank was too low
it would stay full all winter and if it was too high it would virtually be empty all summer.
Steady state model runs could not provide the kind of insight needed to site the tank.

With the help of numerous EPS runs, the author was able to determine a workable
bottom and overflow elevation of the tank, locate control valves and set the timing of the
control valves to insure that the tank water level fluctuated over the desired range while
maintaining adequate pressure at the high points and not overpressurizing the low points.

Use of graphical output made it possible to quickly study the range of alternatives and
provided an excellent means of presenting results to other stakeholders.

Summary

EPS models are now so easy to use that they can be applied to solve a wide variety of
problems. Models can simulate any parameter in the system at any time. The key to
effectively using results is understanding what graphs to call up and what to look for in
them.

About the author. The author is Engineering Manager for Pennsylvania American Water
Company, 20 E. Union St., Wilkes-Barre, PA 18701 and can be reached at
twalski@pawc.com. The figures used in this paper were taken from work prepared by
the author in Essential Hydraulics and Hydrology published by Haestad Methods, Inc.
with the permission of Haestad Methods, Inc..
 (hrl

Figure 1. Effect of Pump Capacity on Tank Level

.-_- .:
,....I .. :. ., 1

E?bct of Main Sue

rdraulic Grade v?nwn~ Time VW

10 0 20.0 30.0 40 0 50 0 60 0 70 0 80 0
Tsme
(hrl

Figure 2. Effect of Main Size on Tank Water Level

 Atternsttve Tank Sues
Hydrauk Grade “ar-ymg Twne

00 so 100 1s 0 20 0 2s 0 30 0 3s 0 40 0
Time
W)

Figure 3. Effect of Tank Volume on Fluctuations

Tank. 30 R tank

---tBaw
---*-- T&LOW

270 0

265.0 I I I I I 1
00 10 0 20 0 30 0 40 0 so 0 60 0 70 0 80 0
Tome
(ha

Figure 4. Effect of Tank Overflow on Water Level Fluctuations

 Junction. J-3
Presrure varymp Trims
70 0
I I I I

001 I I I I I I I 1
00 10 0 20 0 30 0 40 0 50 0 60 0 70 0 60 0
Time
VW

Figure 5. Analyzing T;ow Pressure Problems

60.0.

Figure 6. Evaluating PRV Operation

 --
Tank 30 R tanh
Hydrauhc Qrade vsrymg Tme
299 0

296 0

295 0
00 10 0 20 0 30 0 40 0 50 0 60 0 70 0 80 0

-- Figure 7. Analyzing Frequent Pump Cycling

, .., .,.. *E;/:.$z<.; :y; ” I, . . . ,<< , .:.. :.: <: _ .:... ,. .,

Junction. J-2
Prersure valylnp Time
- 1hste.Q
r\

72 0

Time
OW

Figure 8. High Pressures during Tank Outage

 295 0

290 0
8
5 265 0
0 P
3’
j 260 0

275 0

270 0

265 0
00 50 10 0 150 20 0 25 0
Time
(W
~I ..... ,, ., :<.--

Figure 9. Tank Level Fluctuations during Fire

100 150 20 0 2s 0
Time
V-4

Figure 10. Tank Level Fluctuations during Pipe Break