Integrating Water Systems – Boxall & Maksimović (eds)

© 2010 Taylor & Francis Group, London, ISBN 978-0-415-54851-9

Improving energy efficiency of pumping systems through

real-time scheduling systems

L.K Reynolds
Mouchel Group Ltd.

S. Bunn
Derceto (NZ) Ltd.

ABSTRACT: Optimisation of water supply systems has been the subject of many papers over the past 20
years of CCWI conferences. Most papers have proposed various methods for off-line optimisation of complex
networks to take advantage of low-priced energy tariffs, whereby the cost saving is achieved by rescheduling
as much of the pumping load as possible to occur at night when the energy tariff is lowest. Whilst cost saving
through time-of-use scheduling is an attractive opportunity for water companies, it does not save actual energy,
it simply redistributes its use; consequently, there is no saving in carbon emissions. The only way the system can
be operated more efficiently and hence save carbon as well as energy cost, is if the pumps themselves can be
operated more efficiently.
The paper presents results following implementation of a closed-loop, real-time optimiser to four actual
water supply systems in US. The data shows that by measuring the efficiency of each pump in real-time and
ensuring that each pump in the system is positioned as close as possible to its best-efficiency point, significant
improvements in average efficiency of between 6–9% can be made in addition to the 10% time of use savings.
This in turn translates into real reductions of the carbon footprint.

1 INTRODUCTION 1.2 Water supply optimisation

There has been a large amount of work done within
1.1 Energy use in water supply
the academic community to develop a wide range of
The UK water industry uses approximately 3% of the different optimisation methods for water supply sys-
total electricity consumption, or 4,400 GWh annually tems. Ever since the first UK CCWI conference in
costing some £200 million. This energy causes around the mid eighties, many different techniques have been
1.9 million tones of carbon dioxide to be emitted to the presented claiming impressive theoretical potential in
atmosphere. The majority of the energy use is for lift- terms of speed or financial savings. Unfortunately, few
ing water from river and groundwater sources (26%) have ever been implemented in practice for a variety
into the treatment plant (30%) and then pumping it to of reasons:
service reservoirs (43%) so that the water company
can supply its customers. a) Robustness: There has been a perception that whilst
Historically, whilst energy has been recognised as the algorithms may find an optimal solution, they
a major operating cost, the efficiency of the pump- are not robust to the practicalities of real plant
ing operation has been accepted as a given and apart failures and changes in customer demand.
from addressing efficiency during pump selection and b) Speed: in order to meet changing demand and
pump station design, the overall operational energy equipment availability, decisions need to be made
efficiency has not been of great concern. quickly to adapt to new conditions.
This situation is changing very rapidly, the need to c) Reliability: the models which have been developed
efficiently manage energy is of particular importance require measurements from the actual sensors and
to society as we embark on a low carbon economy. output signals to actuators in the field carried over
Even a relatively small percentage of energy saved, many different forms of telemetry media. The data
by pumping systems, as part of fine tuning of water and signals received are noisy and unreliable, any
supply operations can have a significant impact on errors in the field measurements will result in inac-
green house gas emissions, energy costs, and the curate forecasts of demand or status and hence
organisations’ bottom line profits. wrong decisions.

325
d) Accuracy: The underlying models have to capture handle several hundreds of pumps and control-valves,
the exact form of the real system and all its con- multiple water treatment plant and many demand
figurations. This requires a systematic and detailed zones.
analysis and representation of the real-system in
the model. There is no room for approximations or
simplification. 2 PUMPS AND PUMP EFFICIENCY
e) Confidence: there has been a reluctance to close-
the-loop, i.e. take the human operator out of the 2.1 Theory of pumps
loop, because of the critical importance of the
water supply system. As a result, where open-loop Most pumps used for municipal water supply are of
scheduling systems have been applied, the opera- the centrifugal type driven by electric motors. They
tor is applying his own understanding of the risk of have complex and unique head/flow characteristics
how the system should be operating as a filter of the and associated energy efficiency curves.
outputs of the optimized schedules. Inevitably, this Every pump is designed and manufactured to oper-
filter is unpredictable and changes the performance ate most efficiently at a specific “duty point.” The
of the model. duty point is the point at which the pump produces
the required flow (Q) at the required head (H) which
Most of the academic approaches have focused on it has to overcome.
improving the speed of the algorithm and recently have Pump performance is subject to fixed and transient
adopted GA techniques because of the long solve times head/pressures that the pump encounters in the pump-
and huge dimensions of the possible options. ing pipe network in which it operates. The impact of
Figure 1 shows a solution which has been able to all these fixed and transient head/pressures, present a
close the loop and provide real-time optimisation of complex and challenging case for designers, of pump-
large and complex water supply systems. Aquadapt ing systems, to arrive at an optimal and energy effi-
has taken a different approach, it has focused on tech- cient outcome, over the entire range of pump/pumps
niques to improve the robustness of the solution by operation.
ensuring that the interface between the model and the The analysis of pump performance and operational
real-world is engineered to be robust and presents a outcomes is undertaken by utilizing pump charac-
simple intuitive interface to the operator. It connects teristic curves. These curves are generated by pump
directly to telemetry and reads key information on the manufacturers for each specific pump type.
state of the water supply system. The output is in the Figure 2 shows an idealized set of pump curves plot-
form of start/stop and open/close commands which ted on a common X-axis representing flow. In an ideal
are routed directly via the telemetry or SCADA sys- situation, the duty point is arranged to be exactly the
tem to the PLC or RTUs controlling the pumps and point where the energy efficiency of the pump is at
valves in the field. An absolutely critical aspect is the its maximum, in this way the efficiency of the pump
automation of staggered starts and gradual changes of operation will be maximized.
state which ensures robust plant control. By solving In practice, however, the duty point is hardly ever
the optimisation problem quickly(<5mins), it can be located at the best efficiency point for a number of
run many times during the day, which enables it to practical reasons:
adapt quickly to changing demand and availability of
plant and equipment. a) The water demand changes throughout the day and
In a typical installation, the optimizer runs every hence the pump moves up and down the flow-axis.
half hour and produces an optimal pump and valve b) The system pressure will also change, more users
schedule for up to 48 hours ahead. The system can will reduce the back pressure in the network causing
the pump to move along the head-axis. (The flow

Figure 1. Schematic of a real-time, closed-loop water Figure 2. Idealised set of pump curves. Best Efficiency
network optimiser. Point ideally at Duty Point.

326
 and pressure relationship is known as the system c) Selecting lowest cost sources of water. The software
curve.) holds the marginal cost of production for each of
c) The design duty has to allow for growth in demand the water treatment plants in the system, it selects
over say a 20 year period and therefore a larger duty the plant with the lowest production cost for each
is specified than will be seen when first commis- half-hour period.
sioned, this moves the duty point to the left and d) Shortest path through the trunk distribution net-
lowers the efficiency. work. The software has a hydraulic model at its core
d) The motor is usually over-sized for the pump, this and is able to find the path through the network with
will cause it to operate at a lower efficiency. the lowest total headloss.
e) The pump is installed in pipe-work in the station e) Pump efficiency improvement. Because the soft-
which will throttle the flow and create additional ware holds the actual pump operating curve which
headloss in the delivery system. The actual pump is calibrated from flow and pressure measure-
curve of the installed pump will be different to the ments read from telemetry, and from the monthly
curve of the pump as it was tested in the factory. energy bill, it is able to balance the complex
f) There will usually be more than 1 pump installed in problem of selecting the combination of pump
a station to achieve the required range of flow duty. settings which delivers the overall lowest oper-
Multiple pumps will be connected into a common ating cost and highest possible efficiency. Each
manifold and will interact with each other. pump will be positioned as close as possible to its
g) The net positive suction head (NPSH) will have a Best-efficiency-point (BEP).
major influence on the duty and efficiency curve of
each pump.
h) Pumps wear as they get older, various studies have 3.2 Calculating pump efficiency improvements
been performed but a reliable study (Ref2) has Aquadapt has been installed in four major water sup-
shown that as a rule of thumb a pump will lose plies in US. It has proven itself in independently
5% of its new efficiency over the first 5 years and audited measurements to reduce energy consumption
then stabilise thereafter. per ML delivered by operating pumps and pump sta-
i) The pump station will often be operating together tions at higher points on their efficiency curve. It holds
with other pumps delivering into the same net- pump and efficiency curve information in the internal
work. Whilst it might be possible to arrange one database and uses this to predict where each pump or
or other pump station to operate at its most effi- combination of pumps is likely to operate on its curve
cient if operating alone, it will be difficult to get if run, and does this for each half hour period for the
all stations operating at their most efficient point next 24 hours. It then uses the resulting calculations
when operating together. of kWh per ML delivered to determine the lowest cost
In summary, selecting and designing the right pump solution. In fact in any energy tariff market where the
and pipe-work is a compromise between cost and per- differential between peak and off peak rates is low
formance which is not often performed effectively, the optimiser typically determines that the lowest cost
with the result that most pumps usually run a long strategy is to run pumps efficiently even in peak peri-
way away from their BEP. ods rather than inefficiently in off-peak periods, if this
is the overall lowest cost solution.
Unfortunately likely efficiency gains are hard to
3 AQUADAPT CALCULATES PUMP predict in a system planning an Aquadapt installation.
EFFICIENCY IN REAL-TIME It has always been easier to measure “before” and
“after” kWh per ML delivered figures for Aquadapt
3.1 How Aquadapt saves money systems, then try and predict the results beforehand.
What is surprising is that the efficiency figures are
Aquadapt Optimisation software saves costs in five quite consistent in all Aquadapt installations. In all
main areas: four systems in the US measurements were made
a) Time-of-use (TOU) load shifting, this is the move- which verified efficiency improvements of between
ment of pumping operations from daytime into 6% and 8.8% overall; with some individual pump sta-
night-time when the energy is lower cost. Most tions demonstrating gains of over 20%. Three of these
pump scheduling systems do this in some form. four systems were independently audited.
b) Peak charges avoidance (TRIAD); a natural result
of TOU optimisation. The software will naturally
3.3 Characteristics of the US water
chose to avoid running pumps during high tariff
supply systems
periods which is when peak charges such as triads1
occur. In order to predict energy efficiency improvements
for a new client, the pumping stations were cate-
1
In UK energy market, a triad is one of the 3 highest half- gorised using the efficiency improvement. It has been
hour demand periods in a year. It is calculated at year end and observed that where a pump operates over a wide range
the charge is applied retrospectively. It can account for up to of its pump curve, or where variable speed drives are
5% of the annual electricity bill if it is not minimised. used, there can be quite considerable efficiency gains.

327
Table 1. Reference sites in US.

Demand
Undertaking Popn .K SR Pumps (ML)

E. Bay MUD, Calif. 660 28 66 480

Washington Maryland 1600 57 81 900
WaterOne, Kansas 570 25 84 400
EMWD, Calif. 630 68 149 450

Figure 4. Efficiency improvement in pumps at EBMUD

before and after introduction of Aquadapt.

Figure 3. Direct relationship between efficiency gap and

potential for improvement.

Account is also taken of the current performance

of the pump station, i.e. newer pumps tend to operate
better than older pumps and may already be operating
well on their curve. On the other hand, older pump
stations have worn pumps and may have system curves
Figure 5. Average 11% improvement in mean of average
quite different to those envisaged when they were built. pump efficiency over 4 years.
Figure 3 shows the relationship between efficiency of
a pump before Aquadapt to the efficiency afterwards,
with pumps divided into efficiency bands. software selecting the right pump or pumps at the right
In Figure 3 above we have categorised the origi- time. This is made possible by Aquadapt’s real-time
nal pump efficiency into ranges, the first group is optimiser being able to deduce the moving system
all pumps that were running at less than 40% origi- curve responding to changing diurnal water demand.
nal efficiency, then all pumps between 40% and 50%, In Figure 5, the performance improvement at
50%–60%, 70% to 80% and above 80%. The Y Axis EBMUD is even more pronounced. Recall that this
shows the efficiency GAIN after Aquadapt and as can is based on 2 years worth of accurate energy accounts
be seen, there is a close to linear relationship. Pumps for the “pre-Aquadapt” figures and 4.5 years worth
that were running at lower efficiencies made the most of ‘post-Aquadapt” energy bills from over 40 pumps.
efficiency improvements (about 11% on average gain These figures were scrutinised by EBMUD in a very
in the first group for example) whereas pumps that detailed review process as part of a performance pay-
were relatively new and performing well, i.e. in the ment scheme and are accepted as correct. In fact
70% to 80% range only showed a ∼4% improvement measuring these benefits is very simple; they are sim-
after Aquadapt. ply the monthly water volumes that flowed through
Figure 4 shows similar data obtained for East Bay the pump station divided by the kWh recorded by the
Municipal Utility District (EBMUD) in Oakland Cali- electricity meter.
fornia. This is based on 6 years of data, two years prior
to Aquadapt and four years post Aquadapt. This dataset
3.4 Variable speed drives
contains over 600 accurately measured monthly energy
and flow data points. Many utilities incorporate variable speed drives into
Figure 4 shows a big improvement in the least effi- their pumping stations to improve energy efficiency.
cient pumps, after the optimizer was commissioned, These pumps can operate over a wide range of flows
no pump runs below 65% efficiency whereas before and pressures. The operating points of these pumps
Aquadapt there were many pumps operating between taken from telemetry have been plotted to create a
55% and 65% and more running between 65% and ‘cloud’ of operating points. Figure 6 shows the manu-
75% efficiency. None were running at better than 75% facturers pump efficiency curve adjusted for different
yet now 80% of pumps and pump stations are achiev- shaft speeds using the affinity laws and which inter-
ing this outcome, all done purely through the optimiser sect the data cloud. The dots along the pump curve line

328
 Figure 7. Variable speed pump best efficiency curve.

characteristic (Black-line) which is perpendicular to

the best efficiency curve. This shows that a fixed head
target pressure is not a good strategy to adopt to achieve
most efficient operation. A better strategy, is to allow
the pump to move up and down within a pressure band
to the point where the efficiency is highest and which
also meets the flow and head requirements, this point
has to balance with the effects of other pumps and
demand which will affect the system head in that part
of the network.

4 CONCLUSIONS

Substantial improvements in pump efficiency can be

achieved through real-time monitoring and optimisa-
tion. Average pump energy efficiency improvements
have ranged from 6% to 8.4% for the existingAquadapt
clients with a few individual pump stations showing
improvements of greater than 20%.
When it comes to variable-speed pumps, often
installed to save energy, the efficiency of pumps sup-
plying variable demand is a complex function which
is not usually taken into account in the design or oper-
ation of the system. Cases have been found where the
efficiency can be as low as 8%.
Figure 6. Variation of operating efficiency for range of Significant increases in the efficiency of inter-
pump speeds. connected pumped systems can only be achieved by
continuous monitoring of the pump and the system
indicate that at any particular fixed pump speed there curve and then optimisation of the whole system. The
is a range of head and flow operating points and hence rewards are impressive and lead to real reduction in
a range of efficiencies that the pump is operating over. the carbon footprint, requiring less MWh to deliver the
Figures 6 shows that the variable speed drive oper- same volume of water. This has enabled four US util-
ates over a very wide range of efficiency from 8% to ities to make practical steps towards meeting realistic
78% efficiency with an average of 47%. If this were carbon reduction targets.
a fixed speed pump we would expect a gain of 14%
after Aquadapt as per the chart for EBMUD. However,
because this is a variable speed drive, the potential REFERENCES
savings potential will be much greater.
1) Study of improving the Energy Efficiency of Pumps:
The optimiser knows exactly how to set the speed, European Commission 2001
pressure or flow setpoint of the pump to have it oper- 2) WRC project: CP348A PUMPS – OPTIMISATION
ate at much more efficient ranges of the curve. Figure AND REPLACEMENT 2009
7 shows the curve of Optimum efficiency (Red-line) 3) Pumping Cost Savings in the Water Supply Industry
is nearly vertical, and compares with a fixed head Seminar, IMechE, May 1989

329