The recent deployment of smart metering networks is opening new opportunities for advancing residential water demand management strategies (WDMS) that rely on a better understanding of users’ consumption behaviors. Recent applications showed that retrieving information on users’ consumption behaviors, along with their explanatory and/or causal factors, is key to spot potential areas for targeting water saving efforts and to design user-tailored WDMS. In this study, we explore the potential of ICT-based systems in supporting the design and implementation of highly customized WDMS. On one side, the collection of consumption data at high spatial and temporal resolutions requires big data analytics and machine learning techniques to characterize typical consumption profiles of the metered population of users. On the other side, ICT solutions and gamifications can be used as effective means for facilitating both users’ engagement and the collection of socio-psychographic users’ information. This latter allows interpreting and improving the extracted profiles, ultimately supporting the customization of WDMS, such as awareness campaigns or personalized recommendations. Our approach is implemented in the SmartH2O platform and demonstrated in a pilot application in Valencia, Spain. Our results show how the analysis of the smart metered consumption data, combined with the information retrieved from an ICT gamified portal, successfully identifies the typical consumption profiles of the metered users and recommends alternative WDMS targeting the different users’ profiles.

1. ICT solutions for highly-customized water
demand management strategies
M. Giuliani, A. Cominola, A. Castelletti, P. Fraternali,
J. Guardiola, J. Barba, M. Pulido-Velazquez, A.E. Rizzoli

2. Residential water demand modeling & management
resolution depends on the installed meter, the logging time can be
shortened without installation of smart meters but simply
increasing the traditional reading frequency by the users. However,
so far only ad-hoc studies systematically collected and analyzed
reconstructing the average flow within the pipe with a resolu-
tion of 0.015 L (Kim et al., 2008).
Ultrasonic sensors (Mori et al., 2004), which estimate the flow
velocity, and then determine the flow rate knowing the pipe
section, by measuring the difference in time between ultrasonic
beams generated by piezoelectric devices and transmitted
within the water flow. The transducers are generally operated in
the range 0.5e2 MHz and allow attaining an average resolution
around 0.0018 L (e.g., Sanderson and Yeung, 2002).
Pressure sensors (Froehlich et al., 2009, 2011), which consist in
steel devices, equipped with an analog-digital converter and a
micro-controller, continuously sampling pressure with a theo-
retical maximum resolution of 2 kHZ. Flow rate is related to the
pressure change generated by the opening/close of the water
devices valves via Poiseuille's Law.
Flow meters (Mayer and DeOreo, 1999), which exploit the water
flow to spin either pistons (mechanic flow meters) or magnets
(magnetic meters) and correlate the number of revolutions or
pulse to the water volume passing through the pipe. Sensing
resolution spans between 34.2 and 72 pulses per liter (i.e., 1
pulse every 0.029 and 0.014 L, respectively) associated to a
logging frequency in the range of 1e10 s (Kowalski and
Marshallsay, 2005; Heinrich, 2007; Willis et al., 2013).
So far, only flow meters and pressure sensors have been
employed in smart meters applications because ultrasonic sensors
are too costly and the use of accelerometers requires an intrusive
Fig. 2. Five-years count of the 134 publications reviewed in this study.
A. Cominola et al. / Environmental Modelling Software 72 (2015) 198e214200
Benefits and challenges of using smart meters for advancing
residential water demand modeling and management: A review
A. Cominola a
, M. Giuliani a
, D. Piga b
, A. Castelletti a, c, *
, A.E. Rizzoli d
a
Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milan, Italy
b
IMT Institute for Advanced Studies Lucca, Lucca, Italy
c
Institute of Environmental Engineering, ETH Zurich, Zurich, Switzerland
d
Istituto Dalle Molle di Studi sull'Intelligenza Artificiale, SUPSI-USI, Lugano, Switzerland
a r t i c l e i n f o
Article history:
Received 2 April 2015
Received in revised form
21 July 2015
Accepted 21 July 2015
Available online xxx
Keywords:
Smart meter
Residential water management
Water demand modeling
Water conservation
a b s t r a c t
Over the last two decades, water smart metering programs have been launched in a number of medium
to large cities worldwide to nearly continuously monitor water consumption at the single household
level. The availability of data at such very high spatial and temporal resolution advanced the ability in
characterizing, modeling, and, ultimately, designing user-oriented residential water demand manage-
ment strategies. Research to date has been focusing on one or more of these aspects but with limited
integration between the specialized methodologies developed so far. This manuscript is the first
comprehensive review of the literature in this quickly evolving water research domain. The paper
contributes a general framework for the classification of residential water demand modeling studies,
which allows revising consolidated approaches, describing emerging trends, and identifying potential
future developments. In particular, the future challenges posed by growing population demands, con-
strained sources of water supply and climate change impacts are expected to require more and more
integrated procedures for effectively supporting residential water demand modeling and management in
several countries across the world.
© 2015 Elsevier Ltd. All rights reserved.
1. Introduction
World's urban population is expected to raise from current
54%e66% in 2050 and to further increase as a consequence of the
unlikely stabilization of human population by the end of the cen-
tury (Gerland et al., 2014). By 2030 the number of mega-cities,
namely cities with more than 10 million inhabitants, will grow
number of people facing water shortage (McDonald et al., 2011b). In
such context, water supply expansion through the construction of
new infrastructures might be an option to escape water stress in
some situations. Yet, geographical or financial limitations largely
restrict such options in most countries (McDonald et al., 2014).
Here, acting on the water demand management side through the
promotion of cost-effective water-saving technologies, revised
Contents lists available at ScienceDirect
Environmental Modelling Software
journal homepage: www.elsevier.com/locate/envsoft
Environmental Modelling Software 72 (2015) 198e214
Benefits and challenges of using smart meters for advancing
residential water demand modeling and management: A review
A. Cominola a
, M. Giuliani a
, D. Piga b
, A. Castelletti a, c, *
, A.E. Rizzoli d
a
Department of Electronics, Information, and Bioengineering, Politecnico di Milano, Milan, Italy
b
IMT Institute for Advanced Studies Lucca, Lucca, Italy
c
Institute of Environmental Engineering, ETH Zurich, Zurich, Switzerland
d
Istituto Dalle Molle di Studi sull'Intelligenza Artificiale, SUPSI-USI, Lugano, Switzerland
a r t i c l e i n f o
Article history:
Received 2 April 2015
Received in revised form
21 July 2015
Accepted 21 July 2015
Available online xxx
Keywords:
Smart meter
Residential water management
Water demand modeling
Water conservation
a b s t r a c t
Over the last two decades, water smart metering programs have been launched in a numb
to large cities worldwide to nearly continuously monitor water consumption at the sin
level. The availability of data at such very high spatial and temporal resolution advanced
characterizing, modeling, and, ultimately, designing user-oriented residential water dem
ment strategies. Research to date has been focusing on one or more of these aspects bu
integration between the specialized methodologies developed so far. This manuscrip
comprehensive review of the literature in this quickly evolving water research doma
contributes a general framework for the classification of residential water demand mod
which allows revising consolidated approaches, describing emerging trends, and identif
future developments. In particular, the future challenges posed by growing population d
strained sources of water supply and climate change impacts are expected to require m
integrated procedures for effectively supporting residential water demand modeling and m
several countries across the world.
© 2015 Elsevier Ltd. All ri
1. Introduction
World's urban population is expected to raise from current
54%e66% in 2050 and to further increase as a consequence of the
unlikely stabilization of human population by the end of the cen-
tury (Gerland et al., 2014). By 2030 the number of mega-cities,
namely cities with more than 10 million inhabitants, will grow
over 40 (UNDESA, 2010). This will boost residential water demand
(Cosgrove and Cosgrove, 2012), which nowadays covers a large
portion of the public drinking water supply worldwide (e.g.,
60e80% in Europe (Collins et al., 2009), 58% in the United States
(Kenny et al., 2009)).
The concentration of the water demands of thousands or mil-
lions of people into small areas will considerably raise the stress on
finite supplies of available freshwater (McDonald et al., 2011a).
Besides, climate and land use change will further increase the
number of people facing water shortage (McDonald et
such context, water supply expansion through the co
new infrastructures might be an option to escape wa
some situations. Yet, geographical or financial limita
restrict such options in most countries (McDonald
Here, acting on the water demand management side
promotion of cost-effective water-saving technolog
economic policies, appropriate national and local regu
education represents an alternative strategy for secu
water supply and reduce water utilities' costs (Gleick
In recent years, a variety of water demand manag
tegies (WDMS) has been applied (for a review, see
Jeffrey, 2006, and references therein). However, the
of these WDMS is often context-specific and strongly
our understanding of the drivers inducing people to
save water (Jorgensen et al., 2009). Models that q
describe how water demand is influenced and varies i
Environmental Modelling Software
journal homepage: www.elsevier.com/locate/envsoft
36%
43%
13%
6%
1%
Analysis of 134 studies over the last 25 years
Cominola et al. (2015), Environmental Modelling Software

3. A general procedure
Low resolution
data
High resolution
data
non-intrusive
metering
intrusive
metering
1. DATA GATHERING
Decision tree
algorithms
Machine learning
algorithms
2. WATER END USES
CHARACTERIZATION
Descriptive
models
Prescriptive
models
behavioural
modelling
multivariate
analysis
3. USER MODELLING
4.PERSONALIZED
WDMS
water consumption at
the appliance-level
billed data
disaggregated
water end uses
water consumption at
the household-level
disaggregated
water end uses
qualitative info on drivers
of water consumption
quantitative prediction
of users’ behaviours
High-resolution smart metered
water consumption data
Models of water
user behavior
quarterly / half yearly basis readings
1 kilolitre (=1m3)
Traditional water meters
Smart meters resolution: 72 pulses/L
(=72k pulses/m3 )
Data logging resolution: 5-10 s interval
Information on time-of-day for consumption
Smart water meters
1000 L | quartely readings
0.013 L | 5-10 sec.
Cominola et al. (2015), Environmental Modelling Software

4. A general procedure
Low resolution
data
High resolution
data
non-intrusive
metering
intrusive
metering
1. DATA GATHERING
Decision tree
algorithms
Machine learning
algorithms
2. WATER END USES
CHARACTERIZATION
Descriptive
models
Prescriptive
models
behavioural
modelling
multivariate
analysis
3. USER MODELLING
4.PERSONALIZED
WDMS
water consumption at
the appliance-level
billed data
disaggregated
water end uses
water consumption at
the household-level
disaggregated
water end uses
qualitative info on drivers
of water consumption
quantitative prediction
of users’ behaviours
Models of water
user behavior
Models of water
user behavior
Water Utility problem
Water Demand Management Strategies:
TECHNOLOGICAL (e.g., water efficient devices)
FINANCIAL (e.g., water price schemes, incentives)
LEGISLATIVE (e.g., water usage restrictions)
OPERATION MAINTENANCE (e.g., leak detection)
EDUCATION (e.g., water awareness campaigns, workshops)
Cominola et al. (2015), Environmental Modelling Software

5. A general procedure
Low resolution
data
High resolution
data
non-intrusive
metering
intrusive
metering
1. DATA GATHERING
Decision tree
algorithms
Machine learning
algorithms
2. WATER END USES
CHARACTERIZATION
Descriptive
models
Prescriptive
models
behavioural
modelling
multivariate
analysis
3. USER MODELLING
4.PERSONALIZED
WDMS
water consumption at
the appliance-level
billed data
disaggregated
water end uses
water consumption at
the household-level
disaggregated
water end uses
qualitative info on drivers
of water consumption
quantitative prediction
of users’ behaviours
Models of water
user behavior
Models of water
user behavior
Characterization of the water demand at the household level, possibly
as determined by natural and socio-psychographic factors as well as by the
users' response to different WDMS.
Cominola et al. (2015), Environmental Modelling Software

6. ICT support

7. Research questions
1) Can big-data analytics support the identification of consumption
profiles?

8. Research questions
1) Can big-data analytics support the identification of consumption
profiles?
2) Can gamification induce behavioral change?

9. SmartH2O Project in Valencia
VALENCIA | ES
EMIVASA water supply utility
2 million customers served
490,000watersmart meters currentlyinstalled
Developmentplan: 650,000 water smartmeters installedbyend2015
http://www.smarth2o-fp7.eu/

10. User profiling via hierarchical clustering
Analysis of observed water consumption to study historical trends, extract consumption profiles, identify promising areas
for demand management options

11. Classification of average daily demand
very high (495 l/d)
high (250 l/d)
medium (130 l/d)
low (30 l/d)
DATASET:
11,000 households
hourly readings
2 years monitoring period

12. Weekly patterns
DATASET:
11,000 households
hourly readings
2 years monitoring period
max in week day max in weekend
very high
(495 l/d)
high
(250 l/d)
medium
(130 l/d)
low
(30 l/d)

13. Hourly patterns - load shapes
DATASET:
11,000 households
hourly readings
2 years monitoring period

14. Most common load shapes
DATASET:
11,000 households
hourly readings
2 years monitoring period

15. Most common load shapes
DATASET:
11,000 households
hourly readings
2 years monitoring period
Zero-consumption days

16. Most common load shapes
DATASET:
11,000 households
hourly readings
2 years monitoring period
Double-peak
consumption pattern

17. Model validation
he observed one, with a very small under- and overestimation for medium and high
12. Empirical cumulative density function of daily consumption of EMIVASA
users.
daily average consumption

18. Model validation
Figure 13. Empirical cumulative density function of daily consumption of EMIVASA
users estimated for weekdays (left panel) and weekends (right panel) separately.
daily average consumption only week days only weekends

19. Gamification for behavioral change
DropTheQuestion
available on

20. Water consumption reduction - preliminary results
historical consumption
consumptionaftersH2Otreatment
-10%

21. Behavioral change - preliminary results
0
20
0
20
40
60
80
0
20
40
6080
100
120
0
20
40
60
80
100
20
40
60
80
100
0
0
20
40
60 80
100
120
1400
20
40
60
80
0
20
low
(30 l/d)
low
(30 l/d)
medium
(130 l/d)
medium
(130 l/d)
high
(250 l/d)
high
(250 l/d)
very high
(495 l/d)
very high
(495 l/d)
After sH2O treatment
History stable behavior
decreasing demand
increasing demand

22. Behavioral change - preliminary results
0
20
0
20
40
60
80
0
20
40
6080
100
120
0
20
40
60
80
100
20
40
60
80
100
0
0
20
40
60 80
100
120
1400
20
40
60
80
0
20
low
(30 l/d)
low
(30 l/d)
medium
(130 l/d)
medium
(130 l/d)
high
(250 l/d)
high
(250 l/d)
very high
(495 l/d)
very high
(495 l/d)
After sH2O treatment
History stable behavior
decreasing demand
increasing demand

23. 0
20
0
20
40
60
80
0
20
40
6080
100
120
0
20
40
60
80
100
20
40
60
80
100
0
0
20
40
60 80
100
120
1400
20
40
60
80
0
20
low
(30 l/d)
low
(30 l/d)
medium
(130 l/d)
medium
(130 l/d)
high
(250 l/d)
high
(250 l/d)
very high
(495 l/d)
very high
(495 l/d)
After sH2O treatment
History stable behavior
decreasing demand
increasing demand
Behavioral change - preliminary results
-10%

24. Take home messages
• The large-scale deployment of smart meters is renovating residential water demand
management
• ICT tools have strong potential for making the most of these new data
• Gamification represents a promising option for raising users’ awareness and induce behavioral
change

25. Next Steps
Combine the consumption profiles with psychographic information about the users for a better interpretation/
characterization of the extracted profiles.
Traditional survey: static snapshot and
limited/expensive scalability
Online gamified survey: dynamic data
collection but limited population’s sample

26. thank you
Matteo Giuliani
matteo.giuliani@polimi.it
http://giuliani.faculty.polimi.it