AGENT-BASED DIVISION OF WATER DISTRIBUTION

SYSTEMS INTO DISTRICT METERED AREAS

J. Izquierdo, M. Herrera, I. Montalvo and R. Pérez-García

CMMF, Universidad Politécnica de Valencia, Cno. de Vera, s/n, 46022, Valencia, Spain

Keywords: Multi-agent, Water supply, Decision support system, Water leaks.

Abstract: Water distribution systems are extremely complex assets. The complexity of these systems is continually

increasing from the point of view of technical management. Most water companies do not have the

necessary global vision of their production and distribution, and lack suitable tools to control and operate

their systems. As a result, an increasing number of projects examining suitable methods to divide water

distribution networks into district-metered areas (DMA) are in progress in many cities worldwide. Division

into DMAs aims at splitting large interconnected distribution network into smaller and virtually independent

networks. Division into DMAs enables action to be taken to improve the control of important aspects of

water distribution, such as leaks and water quality. Due to the complexity of the problem, especially for

large systems, efficient techniques are required. In this paper, we introduce a multi-agent approach that

takes advantage of the distributed nature of water distribution systems. This approach has been applied to a

real network with promising results.

1 INTRODUCTION

Water supply is one of the more recognizable and

important public services contributing to quality of

life. It exhibits a number of characteristics that are

quite different from those of other public services.

Its distribution is irregular, both in temporal and

spatial terms. In addition, its operation can be

analyzed from very different perspectives. As water

is for consumption, aspects related to health need to

be considered: water quality and the appropriate

control measures to maintain quality during

residence time in the network. However, water

systems suffer a number of operational and

environmental conditions, which lead to progressive

and insidious deterioration. There are a variety of

factors involved: loss of pressure, due to increasing

inner roughness of pipes; breakage or cracking of

pipes, caused by corrosion and mechanical and

thermal charges; and loss of water (leaks), due to

pipe breaks and cracks, with their corresponding

economic loss, and third party damage. All these

factors can create a risk of contamination. Because

of the complexity of water systems, mainly due to

the interconnected nature of sources and

consumption points, it is extremely difficult to

balance production and distribution, that is to say, to

control the water supplied and consumed. Division

of the network into DMAs follows a divide-and-

conquer strategy that splits a large and highly

interconnected distribution network into smaller and

virtually independent networks – each supplied by a

pre-fixed number of sources.

Independence can be physically achieved in a

number of ways: by closing valves in existing pipes,

by sectioning existing pipes, by introducing new

pipes that redistribute the flow, etc.

Manageable DMAs will enable action to be

taken to improve the control and management of

important aspects of water distribution, such as

water quality and the intensity and spatial and

temporal distribution of leaks. DMAs will help

reduce unaccounted-for water loss and improve the

water-tightness of the system – thereby saving huge

amounts of water and preserving water quality.

Substantial water savings can be forecasted if

sectorization is enforced systematically.

Real water distribution systems may consist of

thousands of consumption nodes interconnected by

thousands of lines, as well as the necessary elements

to feed the network. These networks are not usually

the result of a unique process of design, but the

consequence of years of anarchic response to

continually rising demands. As a consequence, their

Izquierdo J., Herrera M., Montalvo I. and Pérez-García R. (2009).

AGENT-BASED DIVISION OF WATER DISTRIBUTION SYSTEMS INTO DISTRICT METERED AREAS.

In Proceedings of the 4th International Conference on Software and Data Technologies, pages 83-90

DOI: 10.5220/0002237800830090

 SciTePress

layouts lack a clear structure from a topological

point of view. This fact renders these networks

difficult to understand, control, and manage. In the

case of small networks, simple techniques,

sometimes of a visual character, enable division into

a few DMAs. But this task is unthinkable for very

large networks because their complexity renders the

problem virtually unfeasible. As a consequence, new

algorithmic capabilities, not implicitly contained in

the hydraulic model, would be of great interest.

The main objective of creating DMAs (or

sectorization) is to obtain the distributed and

manageably scaled information necessary to perform

key actions in each sector (AVSA, 2009). These

actions include:

• audit the hydraulic efficiency or NRW (non-

revenue water),

• characterize the demand curve, especially the

night flow,

• quickly detect possible leaks by analyzing the

evolution of the night minimum flow,

• check the results of search campaigns and

repair leaks quickly,

• detect fraud, under-registration, or diverse

errors of measurement,

• reduce maintenance costs,

• plan investments when guiding supply to the

sectors with more NRW.

The procedure to define the hydraulic sectors

implies (Tzatchkov et al., 2006):

1. Obtaining the number of independent sectors

in a network layout. A sector of the network is said

to be independent when it is supplied exclusively

from its own water sources, and is not connected to

other sectors in the network.

2. Obtaining the set of network nodes belonging

to each individual sector.

3. Revising proposed sectorization actions, such

as valve closing or pipe sectioning, in case such

actions may cut the water supply for some parts of

the network.

4. Defining the area served by each water source,

and the contribution of each source to the

consumption of each network node.

The first and third of these four tasks are crucial

for detecting errors in the network layout and in

proposed sectorization decisions. The second task is

essential for water audits, and the fourth task is

important for defining and visualizing any proposed

sectorization.

A District Metered Area (DMA) is a part of the

water distribution network that is hydraulically

isolated, temporally or permanently, and ideally has

just one supply node equipped with a flow meter.

DMAs are small zones of the system and usually

contain between 500 and 3000 service connections.

The concept of DMA management was first

introduced to the UK water industry in the early

1980s (Morrison, 2004), and it has been used as an

instrument to monitor and reduce the level of leaks

in water supply systems. The technique was mainly

developed in Europe, and has been used in Latin

America from the 1990s, while it is less often used

in the United States and Canada. The development

of DMAs has been strongly empirical, being based

on technical experience and with very few scientific

contributions. It is necessary to highlight the

contributions in UKWIR (1999) and IWA (2007).

Recently, some proposals have been presented for a

conceptual and scientific framework – such as

Hunaidi (2005) relative to the periodic acoustic

surveys in a DMA; or Tzatchkov et al. (2006), in

applying graph theory to establish the division of

DMAs.

In this paper, we explore the division of a water

supply system into DMAs by using a multi-agent

approach. Multi-agent techniques have proven to be

highly efficient in the solution of very complex

problems of a distributed nature – an example of

which is shown below. In the water field, in

particular, there has been a tendency in recent years

to include multi-agent techniques as an interesting

alternative for solving complex problems that differ

from the problem addressed in this article. See, for

example, (Izquierdo et al. 2008) on multi-agent

applications in urban hydraulics; (Maturana et al.

2006) on water and waste water control system

architecture; (Kotina et al. 2006) on control systems

for municipal water; (Nichita and Oprea 2007) on

water pollution diagnosis; (Feuillette et al. 2003) on

water demand management for a free access water

table; (Hai-bo et al. 2005) on water quality; (Becu et

al. 2001) on water management at catchment scale;

(Cao et al. 2007) on optimization of water networks;

(Mikulecký et al. 2008) on water management at

river basin scale; and (Hailu and Thoyer 2005) on

allocation of scarce water, among others.

Complex problems, such as the problem

considered in this article, can be resolved using

distributed agents because the agents can handle

combinatorial complexity in a real-time suboptimal

approach (Maturana et al., 2004).

The structure of this paper is as follows. Firstly,

we introduce the agent-based ingredients, then

describe the used implementation, and finally,

present the main results. A conclusions section

closes the paper.

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

2 MULTI-AGENT METAPHOR

In the study of complex systems, computer programs

have played an important role. However, the actual

process of writing software is a complicated

technical task with much room for error. Multi-agent

philosophy adopts a modelling formalism based on a

collection of independent agents interacting through

discrete events. Simulation of discrete interactions

between agents stands in contrast to continuous

system simulations, where simulated phenomena are

quantities in a system of coupled equations.

An agent is any actor in a system: any entity that

can generate events that affect itself and other

agents. In the problem we consider below, agents are

consumption nodes, connecting pipes, supply

sources, ground and underground patches containing

the network; as well as district metered areas, which

are set of nodes, pipes, sources, and patches. Even

the whole network is an agent following specific

scheduled actions. In these last two cases, agent

behaviour is defined by the emergent actions of the

agents they contain.

Agents define the basic objects in the system –

the simulated components. The simulation occurs in

the modelled world itself, and it is frequent to speak

of agents as living in an environment, which, in its

turn, can be an agent itself.

Once agents have been defined and their

relationships established, a schedule of discrete

events for these objects defines a process occurring

over time. Individual actions take place at some

specific time; and time only advances alongside

events scheduled at successive times. A schedule is a

data structure that combines actions in the specific

order in which they should execute. The passage of

time is modelled by the execution of the events in

some sequence. Instructions are given to hundreds or

thousands of independently operating agents. This

makes it possible to explore the connection between

the micro-level behaviour of individuals; and the

macro-level patterns that emerge from the

interaction of many individuals.

A final step consists in observing the model and

recording what is happening. Observers perform

these actions. In most platforms there are also agents

with specific tasks, such as plotting, storing data,

monitoring and displaying certain variables, etc.

Agents should possess the following properties:

autonomy, mobility, reactivity, pro-activity,

adaptability, communicativeness, robustness,

learning ability, task-based orientation, and goal-

based orientation (Lee, 2006).

The agent-based approach is worthwhile

(Wooldridge and Jennings, 1995), (Wooldridge,

2002) when facing:

• open, highly dynamic, variable, poorly

structured, uncertain situations

• where the environment can be seen as a system

of autonomous, cooperative, or competitive entities

• data, control, or expertise is distributed

• the system can be divided into independent

components.

3 IMPLEMENTATION

NetLogo (NetLogo, 2007) is an environment for

developing complex, multi-agent models that evolve

over time. It is possible to create populations of

changing agents in a suitable grid of stable agents.

The evolution of agents can take different forms.

Agents can be created, move, change their

properties, change their behaviour, change their

nature or breed, and even die.

Figure 1: Detail of the network in NetLogo environment.

Our model is created from GIS data defining the

physical and topological network characteristics.

The experimental data was obtained from a GIS

model of a real moderately-sized network that has

been studied by the authors within a joint research

project with an international water company. This

network is a part of a water distribution system in a

Latin American capital.

The area is divided into squares (patches), which

gives some raster format to the environment. Patches

represent the ground (underground) where pipes and

nodes are buried. Figure 1 shows a section of the

network. Patches are used to define areas occupied

by the different divisions that will be created. As a

consequence, this raster structure cannot be observed

in Figure 1. Consumption nodes (small circles) are

agents (turtles) of a certain breed with a number of

associated variables. Among the user-defined

AGENT-BASED DIVISION OF WATER DISTRIBUTION SYSTEMS INTO DISTRICT METERED AREAS

variables, elevation and demand are the most

important. During the process, colour is used to

define the DMA that the agents belong to. Pipes

(grey lines) are links (in the problem under

consideration they are undirected links). Each pipe

connects two different nodes and also has some

associated variables. The main user-defined

variables are diameter and length. Source points

(coloured squares) are another breed of turtles,

whose variables are the average of the demand they

supply and the DMAs they feed. Patches, sources,

nodes, and pipes are spatially fixed agents in the

sense that, obviously, they do not change their

position with time. Instead, they change their

properties, especially colour, and as a result they

eventually belong to one DMA or another. Initially,

the sources, nodes, and pipes are presented in light

grey, since no district structure is available at the

setup.

In this model, the user decides the number of

DMAs to be built. Then, randomly the same number

of source points are selected to be the seeds for the

corresponding districts. Upon setup, these turtles

start a process of probing their neighbouring nodes

and checking the likelihood of their neighbours

being assimilated into the same would-be DMA as

themselves. This likelihood is derived from a

number of tests, which are performed on the basis of

sources, nodes, and pipe properties:

• the total length of the current DMA must be

bounded by minimum and maximum values for the

total length of the set of its members,

• the elevation of a new candidate must be in a

certain range around the average elevation of the

current DMA,

• the total demand of the sector must be between

certain pre-fixed limits,

• the associated sources must be able to meet that

demand,

• the geometrical properties of the area occupied

by a DMA must exhibit certain basic requirements

(connectivity, convexity, etc.),

• other properties.

Nodes and pipes passing these tests are

assimilated to the winning DMA, and the process is

repeated again. The process is able to find good

solutions for the connectivity between DMAs. As a

consequence, the number and location of the closure

valves is optimized for a given layout. In addition,

nodes are assigned to sectors in a remarkably stable

way that further stabilizes during the evolution of

the process. Also, the best partitions can be found

with frequency during different runs of the process.

As a result, by repeating the process a certain

number of times, the engineer can make a final

decision that may, or may not, coincide with any of

the obtained partitions – these being used as a basis

for the decision.

Figure 2: Two stages of the sectorization process.

Neighbouring nodes for every hydraulic sector or

division are explored in each step of the algorithm.

These nodes are given a certain probability of

belonging to a given district. This probability

reflects the difference between the elevation of the

node and the average elevation of the district; and

the difference between its demand and the average

demand of the sector. In this way, the simple greedy

competition based on minimum distance among the

districts is improved, and this adds increased

probabilistic richness to the process. As a result, the

model agents, performing a mixture of individual

and collective actions, can explore good network

sectorization layouts.

From the point of view of Logo programming,

the functions self and myself are essential to define

the list of inputs of the agent (set of agents) to be

interrogated, and the agent (set of agents) making

the enquiry. In other words, nodes are assigned to a

given sector after some dialectic process of

neighbouring enquiries among different sectors

competing for the node (see Figure 2 showing two

stages of the process). In the upper part of the figure,

a number of nodes (coloured) have already been

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

assigned to specific sectors, while others (grey) are

still unassigned. Regarding pipes, grey means

unassigned, and red is used to identify pipes

connecting two different sectors.

to add-to-cluster

ifelse any? link-neighbors with [color != [color] of

myself and shape != "square"]

[ask one-of link-neighbors with [color != [color] of

myself and shape != "square"][

let color1 color

set color [color] of myself

if zoning [splotch]

ask my-links [set color [color] of myself]

let elevation-cluster mean[elevation] of turtles

with [color = [color] of myself]

let demand-cluster mean[demand] of turtles with

[color = [color] of myself]

let resist weight-elevation * ([elevation] of self -

elevation-cluster) + weight-demand *

([demand] of self - demand-cluster)

if random 100 < resist [

set color color1

ask my-links [set color color1]

if zoning [splotch]

build-cluster]

]

[stop ]

end

Figure 3: Assignment procedure.

Figure 3 shows the assignment procedure that

decides the assignment of nodes for a calling sector.

Firstly, neighbouring nodes for the calling sector are

scanned. Then a decision is made about the colour to

be assigned to a neighbouring node. This decision is

performed by weighting the difference between the

node’s elevation and demand with regard to the

average elevation and demand of the calling sector

(bold line in Figure 1). This weighted difference

provides a resistance degree (resist) in terms of the

probability of making a decision about joining the

calling hydraulic sector. The weights influencing

this difference are selected by the user through the

slider called ‘weight-demand’ (see Figure 4). Once a

new configuration has been built, other requisites

(sector size, connectivity, etc.) are checked before

validating the configuration. Borders between

sectors – which can be fuzzy – are identified by red

pipes, showing the location of cut-off valves, which

are used to isolate the sectors during operation.

The natural consequence of this process is that

different DMAs are built and minimal sets of

sectioning lines are identified. Nevertheless, some

nodes may end up disconnected and borders between

sectors may be poorly defined. The main objective is

the identification of DMAs and sectioning lines. The

information about disconnected nodes and overlaps

between nodes and pipes of certain colours with

sectors of a different colour (lower part of Figure 2)

can also be used by the network manager. These

circumstances, which show that the desired balance

is still undergoing some debate regarding

assignations, can be used to detect errors in network

data, propose candidate areas for sensitivity

analyses, and encourage various actions aimed at

improving the layout and/or the topology of the

network.

Figure 4: The interface elements.

Through additional interface elements, the user

can manage the course of the simulation by

changing different parameters (see Figure 4). The

membership probability measurements of a node

with respect to a district depend on elevation and

demand. As stated earlier, the user can modify the

weight of these elements by using the slider labelled

‘weight-demand’. The default for this value is 0.5,

and the sum of the weights corresponding to

elevation and demand is 1.

The user can decide a priori the number of

hydraulic sectors to be built by selecting an option

from the chooser labelled ‘n-clus’. By using switch

‘zoning’ the user can also ask the program to colour

the patches occupied by the different hydraulic

sectors. This option, as well as offering an

interesting visual value, enables the user to decide if

the built districts exhibit good topological properties.

Certain convexity and/or compactness properties are

desirable for the districts. By default (option off), the

different colours for the pipes and nodes make clear

the division of the hydraulic network into districts.

By flipping the switch to ‘on’, patches are coloured

according to the colour of the nodes and pipes they

contain. This option is useful for revealing overlaps

AGENT-BASED DIVISION OF WATER DISTRIBUTION SYSTEMS INTO DISTRICT METERED AREAS

between sectors which, as explained before, can be

used to produce suitable sensitivity analyses.

Finally, those pipes that enable

isolation/communication between two sectors are

represented in red, and provide the engineer with

useful information about the candidate pipes for

sectioning and where to install cut-off valves to

isolate the districts.

Figure 5: Interface including monitors.

The simulation results can be visualized on

screens, plots, and graphs; and data can be stored for

further processing in hydraulic simulation software

and for decision-making support.

Figure 5 presents several displays showing some

of the used parameters, such as the average elevation

of the different sectors and the number of pipes in

the sectors. Of special importance is the display

labelled ‘n-valves’ which shows the number of

sectioning links connecting different sectors; and

corresponds to closing valves for the given partition.

Engineers must take important decisions about the

need to install closing valves in existing pipes, and

about sectioning those pipes, and/or introducing new

pipes that redistribute the flow in more a reasonable

manner.

Our simulation model helps managers

communicate with domain experts, because they can

perform their reasoning by solving modelled

situations. The model is based on simple physical

principles, but its large-scale modification is

straightforward.

4 RESULTS

To show the performance of the presented process,

we present the results for a network (see Figure 6)

fed by three reservoirs and made of 132 lines and

104 consumption nodes; its total length being

9.055km, and the total consumed flowrate

amounting to 47.09l/s.

After 20 runs of the model simulating the

partition of this network into hydraulic sectors, the

configuration shown in Figure 7 was obtained in

80% of the cases. As a result, three sectors have

been obtained that are isolated through 11 cut-off

valves (links shown in red). These sectors have 56,

50, and 17 pipes, respectively. All of these sectors

satisfy the requirements for becoming valid

hydraulic sectors in terms of maximum and

minimum total pipe lengths.

The average elevations for these sectors are 73m

for the brown sector, 69m for the yellow sector, and

again 73m for the green sector. As shown by the

plots in Figure 5, the architecture of the sectors

stabilizes during the simulation. In this case, it is the

evolution of the average sector demand that is

represented. We have chosen this graph since it

shows the main reason why the green sector appears

as a genuine sector, despite the fact that its average

elevation coincides with that of the brown sector.

Figure 6: Layout of the studied network.

ICSOFT 2009 - 4th International Conference on Software and Data Technologies

Finally, it is noteworthy that the validity of the

district division has been checked using EPANET2.

The analyses performed show that the proposed

division effectively cuts the water supply for the

desired parts of the network. Also, the entire

network and individual sectors maintain all their

design requirements. As a consequence, the

proposed division into DMAs is perfectly feasible

and reliable.

5 CONCLUSIONS

The multi-agent metaphor has been used with

success in different areas and is reasonably

applicable in the water management field. In

addition to the traditional centralized architecture of

a single reasoning agent (the computing counterpart

of human decision support), it is possible to use

systems of reasoning agents, or to apply multi-agent

simulations to verify hypotheses about the various

processes in water distribution. Partial

implementations of multi-agent applications are

expected to simplify communication with domain

experts during the process of modelling the expert

knowledge, identifying needs, and summarizing

requirements for final application operation. Among

the various scenarios using multi-agent systems in

the scope of decision support for a water

management company, we have focused on the

division of a network into district metered areas.

Division into DMAs helps the decision-making

process, as well as the implementation of suitable

actions to improve the control of a network and give

solutions for important aspects of water distribution

management – such as leaks and water quality.

Figure 7: Final distribution of hydraulic sectors.

We will be researching various lines of research

in the future. One line of action will be addressed to

exploiting the presented model in a number of ways:

adding new conditions to cluster building; refining

the implemented clusters; and generally improving

the performance of the model. An interesting

improvement would consider starting the process

automatically, instead of waiting for the user to

define a priori the number of sectors, thus

performing the division into an optimum number of

sectors. The model must also be applied to larger

networks. Nevertheless, taking into account that the

considered network is medium-sized and running

times are slow (ranging between 10 and 20 seconds

on a PC with an Intel Core 2 Duo T5500 (1.66 GHz)

processor for the case considered), no added

difficulties are foreseen. Another line of research

will focus on the development of other scenarios for

multi-agent applications in the water supply field,

including aspects related to water quality and other

managerial issues.

ACKNOWLEDGEMENTS

This work has been performed with support from

grants BES-2005-9708 and MAEC-AECI

0000202066, awarded to the second and third

authors, respectively, by the Ministerio de

Educación y Ciencia and the Ministerio de Asuntos

Exteriores y Cooperación of Spain.

REFERENCES

Tzatchkov, V.G., Alcocer-Yamanaka, V.H., Bourguett-

Ortíz, V., 2006. Graph Theory Based Algorithms for

Water Distribution Network Sectorization Projects.

8th Annual Water Distribution Systems Analysis

Symposium, Cincinnati, Ohio, USA, August 27-30.

Izquierdo, J., Montalvo, I., Pérez, R., 2008. Aplicaciones

de la inteligencia colectiva (multiagente) para la

optimización de procesos en hidráulica urbana. Invited

talk at VIII Seminario Iberoamericano – SEREA

Influencia sobre el Cambio Climático, la eficiencia

energética, de operaciones y Sistemas de Seguridad en

el abastecimiento y el drenaje urbano.

Maturana, F.P. et al., 2004. Real time collaborative

intelligent solutions, SMC (2) 2004, 1895-1902.

Maturana, F.P., Kotina, R., Staron, R., Tichý, P., Vrba, P.,

2006. Agent-based water\waste water control system

architecture. IADIS International Conference Applied

Computing.

AGENT-BASED DIVISION OF WATER DISTRIBUTION SYSTEMS INTO DISTRICT METERED AREAS

Nichita, C., Oprea, M., 2007. Water Pollution Diagnosis

with a Multi-Agent Approach. In Proceedings,

Artificial Intelligence and Soft Computing.

Kotina, R., Maturana, F.P., Carnahan, D., 2006. Multi-

agent control system for a municipal water system.

Proceedings of the 5th WSEAS International

Conference on Artificial Intelligence, Knowledge

Engineering and Data Bases, Madrid, Spain, 464-469.

Feuillette S., Bousquet F., Le Goulven P., 2003. SINUSE:

a multi-agent model to negotiate water demand

management on a free access water table.

Environmental Modelling and Software, 18(5), 413-

427.

Hai-bo, L., Guo-chang, G., Jing, S., Yan, F., 2005. Multi-

agent immune recognition of water mine model.

Journal of Marine Science and Application, 4(2), 44-

49.

Becu, N., Perez, P., Walker, A., Barreteau, O., 2001.

CatchScape: An integrated multi-agent model for

simulating water management at the catchment scale, a

northern Thailand case study. In: Ghassemi, F. et al.

(Eds). Integrating Models for Natural Resources

Management Across Disciplines, Issues and Scales.

International Congress on Modelling and Simulation,

Canberra, Australia, 1141-1146.

Cao, K., Feng, X., Ma, H., 2007. Pinch multi-agent

genetic algorithm for optimizing water-using

networks. Computers & Chemical Engineering,

31(12), 1565-1575.

Mikulecký, P., Bodnárová, A., Olševičová, K., Ponce, D.,

Haviger, J., 2008. Application of multi-agent systems

and ambient intelligence approaches in water

management. 13th IWRA World Water Congress,

Montpellier (France).

Hailu, A., Thoyer, S., 2005. Multi-Unit Auctions to

Allocate Water Scarcity Simulating Bidding

Behaviour with Agent Based Models. LAMETA

Working paper 2005-01, EconWPA.

NetLogo homepage, 2007 [online]

http://ccl.northwestern.edu/netlogo/

Wooldridge, M., 2002. An introduction to MultiAgent

Systems. John Wiley & Sons, Chichester, UK.

Wooldridge, M., Jennings, N., 1995. Intelligent agents:

theory and practice. In: The Knowledge Engineering

Review 10(2), 115-152.

Lee, R.S.T., 2006. Fuzzy-Neuro Approach to Agent

Applications: From the AI Perspective to Modern

Ontology. Springer-Verlag.

Aguas de Valencia, (2009). Sectorización [online] URL:

https://www.aguasdevalencia.es/portal/web/Tecnologi

a/Tecnologia/GestionRedes/Sectorizacion.html

[accessed: 11 March 2009]

Morrison, J., 2004. Managing leakage by District Metered

Areas: a practical approach. Water 21, 44-46.

UK Water Industry Research Ltd, 1999. A Manual of

DMA Practice. London: UK Water Industry Research.

IWA Water Loss Task Force, 2007. District Metered

Areas: Guidance Notes. Version [online] URL:

http://www.waterlinks.org/upload_file/8f4be72e-

5027-cbe8-40ed-ee12d1fe6495.pdf, February 2007,

100 pp., [accessed: 11 March 2009]

Hunaidi, O., 2005. Economic comparison of periodic

acoustic surveys and DMA-based leakage

management strategies. Leakage 2005 Conference

Proceedings, Halifax, N.S., Canada.322-336.

ICSOFT 2009 - 4th International Conference on Software and Data Technologies