Towards Development of An Optimization Model To Identify Contamination Source in A Water Distribution Network
Towards Development of An Optimization Model To Identify Contamination Source in A Water Distribution Network
Towards Development of An Optimization Model To Identify Contamination Source in A Water Distribution Network
Article
Towards Development of an Optimization Model
to Identify Contamination Source in a Water
Distribution Network
1, 1,2 1 3
Oluwaseye Samson Adedoja * ID , Yskandar Hamam ID
, Baset Khalaf and Rotimi Sadiku
1 Department of Electrical Engineering/French South African Institute of Technology (F’SATI), Tshwane
University of Technology, Pretoria 0001, South Africa; hamama@tut.ac.za (Y.H.); khalafb@tut.ac.za (B.K.)
2 École Supérieure d’Ingénieurs en Électrotechnique et Électronique, Cité Descartes, 2 Boulevard Blaise
Pascal, 93160 Noisy-le-Grand, Paris, France
3 Department of Chemical Engineering, Tshwane University of Technology, Pretoria 0001, South
Africa; sadikur@tut.ac.za
* Correspondence: princeturn205@yahoo.com; Tel.: +27-632-453-598
Abstract: Protection of the water system is paramount due to the negative consequences of
contaminated water on the public health. Water resources are one of the critical infrastructures that
must be preserved from deliberate and accidental attacks. Water qualities are examined at the
treatment plant. However, its quality can substantially be contaminated during transportation from
the plant to the consumers’ taps. Contamination in water distribution networks (WDNs) is a danger
that can have severe consequences on public health as well as an economic and social instability.
Water distribution networks are immensely susceptible to deliberate or accidental attacks due to
the complex nature of the system. Hence, contamination source identification (CSI) is a topical
issue in water distribution systems that require immediate attention of researchers in order to
protect mankind from the adverse effect of consuming contaminated water. Usually, a contaminant
event can be detected by the water quality monitoring sensors or the contaminant warning system
(CWS) installed on the network. Nevertheless, how to derive the source of the contamination from
the collected information is a difficult task that must be tackled in order to evaluate the spread of
the contamination and for immediate remedial strategies. In the past two decades, considerable
efforts and advancement have been made by researchers applying various techniques in order to
locate the source of the contamination in WDNs. Each of the techniques has certain limitations and
applicability as reported in the literature. This paper presents a comprehensive review of the
existing techniques with emphasis on their importance and technical challenges. Despite a series
of investigations in this domain, the field is yet to be unified. Hence, open research areas are still
available to explore. Consequently, improvement on the existing techniques is necessary and
hereby suggested. More importantly, practical application of these techniques offer a major
research gap that must be addressed.
Keywords: water distribution networks; contamination source identification; contamination warning system
1. Introduction
One of the most important environmental challenges in the 21st century is the protection of a water
distribution network that supplies clean water to consumers around the globe. The drinking water system
is a physical, cultural, social, political and economic resource crucial to human health and well-being.
Provision of drinking water is a human right and significant to the existence all humans. Hence, the water
distribution system is one of the critical infrastructures and essential to the society [1].
Water distribution networks (WDNs) are complex systems that are composed of spatially interrelated
components [2,3]. Typical WDNs may be described as a collection of nodes and links that transport
water from the main supply to the consumers. Despite the fact that some of the components are
buried, the system can be accessed where human imposition is possible. Thus, it is vulnerable to
accidental and deliberate attack that can have severe consequences for public health, besides
economic and social instability [4–7]. Generally, water quality is examined at the treatment plant, but
the water may be contaminated during transportation to the users. Intrusion into the network may
occur through: storage tanks, nodes, pipe leakages, and intersections [8]. The economic and social
consequences of such occurrences may be grim. The study by Kenzie et al. [9] and Corso et al. [10]
reported the significant effect of a transported infection through the water supply system in
Milwaukee (USA) that engender 403,000 users and subsequently hospitalised with an estimated bill
of about USD 96.2 million. The recent study by Cooper et al. [11] accounted for the consequences
of accidental pollution of chemical in a water distribution network in Virgina where over 300,000
users were affected. Investigations have shown that attacks on the water distribution networks are
real, as it has happened before and can happen again.
In order to minimise the risks to the society, two preventive operations are required: enhancing the
physical security of the system and installment of water quality monitoring sensors. If all nodes in the
network can be systematically surveyed, then a moderate level of safety can be attained. Regrettably, it is
impossible to install sensors at every node in the network due to the high cost of procurement and budget
constraints. However, a reasonable number of investigations [12–30] have concentrated on the sensor
placement techniques to meet diverse objectives. Nonetheless, when a contaminant event is recognised
by the monitoring sensors, it is necessary to accurately identify the source of the contamination for
immediate action in order to safeguard the environment and the mankind. Contamination source
identification (CSI) characterises contaminant location, the time of injection and its concentrations in the
entire network. A derivation of this information from the data obtained from the water quality monitoring
stations is a complex task that must be addressed for several purposes. A contaminant transport model is
an absolute process and treated as ill-posed because its resolution does not meet stability, uniqueness
and existence. In reality, the resolution of existence can be presumed to be fulfilled, since the intrusion
had to be initiated from somewhere. However, the two remaining resolutions are not fulfilled [ 31]. This
produces discontinuous dependence on data and response to the measurement inaccuracy. Much earlier
investigations have made a concentrated effort to address the CSI challenge by several approaches,
which includes particle backtracking by Laird et al. [32] and De Santics et al. [33], machine learning by
Wang et al. [34], and data mining by Huang et al. [35]. Recently, a survey on monitoring, modelling and
optimisation techniques was discussed by Tsitsifli and Kanakoudis [36] and Kanakoudis et al. [37]. The
study by Deng and Wang [38] discussed water quality data analysis based on data mining. Despite
several advancements made by scholars, there is no consensus among them, which shows that the field
is yet to be harmonised.
In this paper, a comprehensive review of the existing approaches for the contamination source
identification in the water distribution network is presented. To the best of our knowledge, there is no
existing publication of similar studies in this domain, making it unique. Hence, it pioneers the review study
in this field upon which subsequent investigations stand. The uniqueness of this study can be attributed
to the following: (1) it accounts for diverse approaches that have been proposed to address the CSI
problem; (2) it outlines various technical associated challenges; (3) suggests possible solutions in order
to develop an improved method and (4) recommends areas of future research directions. The remainder
of this paper is presented as follows: Section 2 elucidates the contamination source identification
problem, whereas Section 3 details the water quality models. In Section 4, solution approaches to pipe
networks was discussed. Quality modelling approaches are depicted in Sections 5 and 6 lists the
available simulation tools. In Section 7, solution approaches to the source identification problems are
presented while Section 8 gives a summary of the existing approaches.
Water 2018, 10, 579
Section 9 details the technical challenges, suggested possible solutions and the future research direction.
Finally, conclusions are drawn in Section 10.
Figure 1. Sample of water distribution network with contaminant injection adapted from [48].
The study by Rossman and Boulos [58] broadly classified the water models into four (4) classes and
compared the speed and accuracy by using a solver developed by Rossman [47]. The four (4)
classes consist two types of Lagrangian and two of Eulerians methodologies. Lagrangian
approaches trace the movement of flows in a pipe and strategically situate flow volume boundaries
where significant changes of concentration can be monitored. Time-driven methods (TDMs) of
Lagrangians used equal-sized time steps like the Eulerian approaches. Liou and Kroon [59]
described an application of this approach in their study. Lagrangian event-driven models (EDMs)
utilised time steps in varying length. Boulus et al. [60] utilised the (EDMs) approach to deduce
optimal segmentation procedure with negligible parts to process the simulation. In relation to the
investigation by Boulos et al. [61], the result permits more dynamic water modelling in reacting to the
shape and length of the network. The Eulerian method partitioned pipes into evenly spaced grid
points for the Eulerian finite-difference method (FDM) and equal size volume for Eulerian discrete-
volume method (DVM). The work of Islam and Chaudhry [62] is an example of FDM while
applications of DVM are reported by Clark et al. [4,5], Grayman et al. [63], and Rossman et al.
[64,65]. Overview of the existing water quality models is presented in Table 1.
Figure 2. Various approaches to the solution of pipe network analysis adapted from [78].
Water 2018, 10, 579
Dynamic models can be spatially categorised as either Eulerian or Lagrangian and temporally as
time-driven or event-driven. In Eulerian models, pipes are divided into fixed control volumes and changes
in the movement of water are recorded, whereas Lagrangian models traces the variation in a series of
discrete parcel as water flow within the pipe. Time-driven models update the state of the network at fixed
time intervals while event driven models adjust the state in the network only when variation is observed
such as when a new parcel of water gets to the extreme of the pipe [ 58]. The article by Rossman and
Boulos [58] made a comparison between four numerical methods (two Eulerian and two Lagrangian
approaches considering both time-driven and event-driven approaches) for modelling the water quality
behaviour in water distribution system. The results of the investigation show that the Lagrangian methods
were more efficient for simulating chemical transport and for modelling the water
Water 2018, 10, 579
age, while the Eulerian methods proved to be memory efficient. In 2004, Munavalli and Kumar [ 70]
compared Lagrangian time-driven methods and event-driven methods for deviation of concentration and
water quality time step. The proposed hybrid method improved the accuracy of the methods. The authors
added that the time-driven methods were affected by both concentration tolerance and water quality time
step while event-driven methods relied on the concentration tolerance. The proposed hybrid model
verified less sensitivity to the parameters and demand a sensible computational effort. The numerical
discretisation employed to model the advection in the water networks presents a good solution for
Gaussian initial distribution, but generated artificial diffusion when steep gradients are simulated.
Simplicity and modest computational effort have made the finite different methods to be more suitable for
one-dimensional problems. Explicit finite-difference techniques are generally simplified programming, with
the time step size usually restricted by the courant stability condition. The majority of implicit finite-
difference methods are unconditionally stable; however, they are significantly more complex and require
extra parameters [87]. Investigation of the application of seven finite-difference methods and one
polynomial interpolation scheme for the solution of the transport equation for both Gaussian and non-
Gaussian initial distributions were presented by Islam and Chandhry [88]. The results obtained by each
method were compared with the exact solution. Almost every method gave reasonable results for a
Gaussian initial distribution. However, for steep gradient concentrations, some of the methods produced
high oscillations. The third and fourth-order methods generated the best results for this scenario, but
required additional computation time and the programming became more complex. Substantial
investigation of the models presented so far employed the extended-period simulation (EPS) approach to
address the flow conditions in the networks. Development of a dynamic model to estimate the spatial and
temporal distribution under slowly varying flow condition was discussed by Islam and Chandhry [ 62]. In
this method, flow propagation was first computed followed by a simulation of a substances simulation with
separation of advection and dispersion. However, the results were not satisfactory when the flow became
more unstable and backward flow materialised. Ozdemir and Ger [82] evaluated the consequences of
daytime operation to the night. They argued that dispersion should be accounted for and they came up
with the development 2D convective dispersive algorithm. The developed algorithm was compared to the
experiment and results verified that the algorithm failed to enhance the output. In 2002, Ozdemir and
Ucak [84] proposed a solution to assess the deterioration of chlorine in the water network by applying a
two-way approach to model the transportation of chlorine. The deterioration calculation in a single pipe
involved the bulk-flow reaction, radial diffusion and pipe wall reaction of chlorine.
An input–output algorithm proposed by Zieroff and Polycarpou [89] expressed the chlorine
absorption at a specific pipe junction and time as a weighted average of exponentially decayed values of
the concentrations at all adjacent upstream junctions. The algorithm traces all routes from the supply
stations through which each measured point to account for the reaction rate related to chlorine
decomposition at the pipe wall to allow modification in order to enhance the estimation of chlorine
absorption and increase the model accuracy. A particle backtracking algorithm (PBA) employed by Shang
and Uber [90] is an easy and systematic class of the input-output model developed by Zieroff and
Polycarpou [89]. Besides that, the model was expanded to permit multiple quality data. The main
advantage of this algorithm is the capability to analyse the specific paths and its attributes since they are
made explicit by the scheme. Analysis of the set of governing equations by Al-Zahrani and Moied [91]
described the reaction–diffusion, the convection system for a homogeneous tubular reactor that
considered both steady-state and dynamic systems by employing various standard reduction
approaches. Results signified that, for cases dominated by convection and reaction, high-resolution
discretisation techniques oscillations were efficient for tracking steep moving fronts. Different types and
order of finite difference methods were reported to model the accumulation and the convection
derivatives. First-order finite difference methods result in univalent and steady conditions with the
inaccurate solution of the systems. Unlike first order, higher order techniques are less disruptive, but
susceptible to uncertainties as reported by Wang and Hutter [92]. Remarkable
Water 2018, 10, 579
development has been recorded to enhance the accuracy of the methods. One standard method to
minimise the unseen vibration close to discontinuities is by applying numerical propagation to the
system. The numerical propagation can be adjusted widely close to discontinuities in order to
minimise the vibrations in another place to keep accuracy constant. Various other approaches have
been developed with this approach. Total variation diminishing (TVD) schemes and flux and slope-
limiter approaches, such as: essentially non-oscillatory (ENO), weighted essentially non-oscillatory
(WENO) schemes and Superbee flux corrector, are few examples of these methods as reported by
Alhumaizi [93]. These authors made a comparison between several finite different models proposed
to address the convection–reaction issue. The results demonstrated that first-order schemes
demanded large grid numbers to enhance the accuracy of the solution, which may not be suitable
for large distribution systems, while higher-order methods can cause high oscillations near
discontinuities. High resolution TVD, WENO schemes and the Superbee flux corrector methods
were tested and proved to be accurate for solving cases with steep concentration profiles. In 2007,
Alhumaizi [94] analysed the strengths and limitations of five flux-limiters to numerically simulate the
behavior of a homogeneous tubular reactor with steep moving fronts. All methods were successful
to solve the cases with steep concentration profiles without giving negative concentrations. The
Superbee limiter was found to be the fastest scheme for simulating the sharp front of the model for
all scenarios. According to Hallam [95], the chlorine decay rates in water distribution networks for
bulk and wall demands should be modelled separately because it depends on various factors. The
author developed a methodology for the laboratory determination of the wall decay rate. The results
revealed that the wall decay rates were restrained by chlorine transport for pipe materials with high
reactivity and by the pipe material characteristics for pipe materials with low reactivity. In general,
water velocity increased wall decay rates, although less statistical confidence was recorded for low
reactivity pipes. A moderate biofilm coating did not affect the wall decay rate for low reactivity pipes.
An inverse model was presented by Munavalli and Kumar [96] to determine the water quality
parameters for both bulk and pipe wall reactions and the source strength of chlorine necessary to
maintain the specified chlorine residual at a target node. The extension of this method with dynamic
hydraulic behavior was discussed by Munavalli and Kumar [97]. The application of the method to a
real-life scenario demonstrated that the quantity, quality and location of measurement nodes played
an important role in the estimation of parameters.The study by Yang et al. [98] defined contaminant
chlorine reactions taking place during the transport of water in pipes. The essence was to
characterise the hydraulic dispersion of non-reactive chemicals in order to improve the detection of
contaminants using water quality sensors and to establish a model for predicting the fate and
transport of a slug of a reactive contaminant. They performed a series of experiments in a pilot scale
network and made a comparison with the results of the simulations. The results showed good
agreement and helped to deduce that the residual chlorine loss curve and its geometry are useful
tools to identify the presence of a contaminant slug and infer its reactive properties in adaptative
contamination detections. In water systems, sensitivity analyses are used to examine the estimation
of physical parameters by empirical techniques. These derived analyses are highly dependent on
the unsteady advection reaction equations for quality modelling [99]. The study by Gancel et al. [99]
proposed a splitting method to simultaneously solve the sensitivities and the advection–reaction
equation that describes the quality modelling. The advection term is solved with an Eulerian scheme
using a TVD criterion and the ordinary differential equations with reaction are resolved using an
implicit third-order Runge–Kutta scheme. The work by Mansour-Rezaei et al. [66] coupled a
Lagrangian-based transport model with a Eulerian-based transient hydraulic model to examine the
contaminant infringement and its spread in a water distribution system. The Lagrangian-based contaminant
transport model minimised the simulation period and avoided the numerical dissolution and dispersion
inaccuracy that are peculiar to the Eulerian-based approach. However, assessment of contaminant
infringement was carried out using the proposed approach to ascertain its capability to reveal the
contaminant infringement and dispersion in the network.
Water 2018, 10, 579
6.1. EPANET
EPANET is one of the most popular software programs in water distribution networks. It can
execute water quality movement and propagation in a piping system. Its performance includes
traces of flow in pipes, pressure at each junction and chemical concentration within the whole
network during an extended simulation analysis. It permits adjustment of input information and
presents results in various formats. It has the capability to analyse both bulk flow and pipe wall in the
piping networks. It can model the water age and tracks the movement of reactive materials over the
time. The hydraulic head lost by the flow of water due to the friction in a pipe and the pipe walls can
be computed by using either: Hazen-Williams, Darcy-Weisbach or Chezy-Manning developed by
Rossman [47]. This package is freely available.
EPANET MSX
There are several other types of water quality problems that cannot be accurately modelled by
using the single species capabilities by Rossman et al. [47]. EPANET-MSX is an expansion of
EPANET that allow complex model formulation of both for bulk and wall pipe reaction. It is also
available as freeware where various chemical reaction models can be executed. The following
studies [76,100–105] have used EPANET for their investigations in water distribution systems.
6.2. PORTEAU
PORTEAU is a tool employed to model the behavior of looped main networks distributing water
under pressure. It provides a decision-making tool for designing and managing drinking water supply
networks. The general principle is simple. It consists of creating a diagram of the particular network,
which is composed of pipes and nodes. The data of the different components of the system will be
inserted to ensure that the complete network infrastructure and conditions are well configured close
to reality. For water quality related cases, three types of results are generated: the concentration of
a product, the age of the water and the origin of the water. The results can be displayed either in the
form of a table at each of the time steps (5 min to 1 h) for the whole network, or in the form of a table
over the whole day by pipe or by node, or in the form of variation curves over the day of pipe and by
node on the network diagram. This software is available as freeware.
6.3. Piccolo
This package has the capability to simulate propagation problems in a piping network.
Simulations can be run for steady state or dynamic conditions. A calibrated model can be used for
master planning, case studies and water quality studies, such as influenced areas, chlorination,
origin of water and blended sources.
to the EPANET algorithm for solving the hydraulics, the main difference is that, when the change in
flow from the previous iteration to the next is less than one tenth of one percent of the overall
propagation, the network is considered solved.
6.5. WaterGEMS
WaterGEMS is a multi-platform environment simulation package that can handle the water
distribution system analysis. It has the capabilities to analyse pipes and can build hydraulic models.
It can also simulate the water network in real time and can identify water loss. It provides an
environment for engineers in order to analyse, design, and optimise in water distribution systems.
6.6. H2ONET
Spanning all platforms, from AutoCAD to ArcGIS to the Web, H2ONET products are stand-alone
software. It is a robust package that can model a complete water distribution network. Its capabilities
include water quality modelling, real-time simulation and control, fire flow and energy management. The
package can also effectively used to analyse pressurised sewer collection systems. It has interfaces to
connect with online Supervisory Control and Data Acquisition (SCADA) automatically.
An optimisation approach is often taken to solve an event detection problem [ 46]. The spread of
contamination in the network can be resolved by a simulation method if the contaminant intrusion is
known by assumption. Modelling the output condition from a known input parameter is referred to as a
forward problem. Contrarily, optimisation techniques attempt to extract the unknown parameters that
produce partly known output conditions. Location of the source of contamination and other features can
be derived from the information collected by the water quality monitoring sensors. This type of challenge
is fundamentally ill-conditioned, which generates distinctive complications that were not obtained in the
forward problem. The CSI problem is commonly expressed as an inverse problem attempting to locate
the source based on the information collected from the water quality monitoring stations. A quadratic
programming (QP) method was presented by Laird et al. [32] and van Bloeman et al. [112] to address the
inverse problem. The optimisation problem can be represented as:
Water 2018, 10, 579
N T
est obs 2
MinimiseE = åå(Ci (t) Ci (t)) (1)
i i
where E is the error, C iest is the estimated contaminant concentration, and C iobs(t) is the observed
contaminant concentration. The time step is t and i is the node index of a water quality sensor. The
indicated number of time steps T and the number of sensors in the network, N. The variation between
estimated and observed pollution profile is calculated for all sensors in the network at every simulated
time step. The objective was to minimise the variation between the measured data and the computed
values of the developed model and utilise regularization to detect the family probable solutions.
Experimental results from the network examined showed that the performance of the algorithm was
affected by changing the location of internal node. Even though the formulation exhibits the possibility to
identify the source location, the excessive computational effort associated with it is a major challenge.
Hence, it is practically impossible to apply this approach to a large distribution network. Exact gradient
calculation through direct or adjoint sensitivities will greatly improve the performance of the algorithm.
Due to the shortcoming of direct linear solvers, it becomes relatively difficult to solve the entire network
problem for increasing large networks. Thus, Laird et al. [113] formulated a dynamic optimisation method
based on a sub-domain approach to reduce the excessive computational time. Analysis of the network
examined showed that the sub-domain approach reduces the computational time and also improves the
quality of the inversion solution. Selection of a sub-domain reduces the amount of the available
information for the optimisation and also reduces the computational time. Caution must be taken while
selecting the sub-domain in order not to exclude the important information of the inversion problem.
However, it is important to identify the optimal sensor locations to minimise solution of non-uniqueness
and reduce the identification time. Most importantly, further investigation of more advanced techniques for
selecting this sub-domain is inconsequential. A particle backtracking algorithm introduced by Zieroff and
Polycarpou [89] and extended by Shang and Uber [90] was presented for breaking down the water quality
model down to its essential elements, calculating an impact coefficient that describes the concentration of
selected outputs as functions of network sources and tank concentrations. This algorithm reformulates
the pipe constraints of the water quality models by reducing the size of the discretised problem and
provides a straightforward mathematical representation of the pipe boundary concentrations. The authors
also identified some important areas of future work such as the determination of optimal sensor location,
the testing of the reliability of the formulation against sensor failure or noise in flow rates or sensor
measurements and the correct description of the reaction rate in pipes. Some results were presented in a
real municipal water distribution system to demonstrate the effectiveness of the proposed simulation. A
source inversion problem developed by Preis and Ostfeld [114] discussed the use of a coupled model
tree-linear programming approach. Primarily, the model trees were formulated from a learning process by
applying tool developed by Rossman [47]. Thereafter, the model trees, linear rule configuration was
combined with a linear programming design to unravel the inverse issue. Also, Preis and Ostfeld [ 115]
suggested a method that linked the tool developed by Rossman et al. [47] with a generic algorithm (GA) to
solve inverse problem. Contaminant events were simulated and the GA regulates the features of injection of
the contamination with the purpose of minimising the variation between the measured data and the
simulated contamination concentrations. Although, the model employed an optimisation heuristic search
technique, it is important to know that such approach is substantive only if a systematic search technique is
not viable. Simplicity of system size and provision of solutions to problem of uniqueness is two major
contributions of this study. However, one major notable limitation of this work is the solution to the
excessive computational burden especially when uncertainty is incorporated. In reality, implementation of
the algorithm will possibly require parallel computing. Additionally, the algorithm assumed that the flows in
the pipe are known and monitoring stations are perfect. This may not be practicable as little information
about the consumption and pressure demand are known as well as utilisation of imperfect sensors.
Investigation on the effect of uncertainties in the sensor measurements
Water 2018, 10, 579
was carried out and presented using a modified generic algorithm scheme by Preis and Ostfeld [116] and
Prasad and Park [117]. The developed model was implemented by using three types of sensors: perfect
sensors, sensors transmitting fuzzy measured information and sensors indicating only a contamination
presence. Two illustrations of increasing complexity were presented to validate the capability of the
developed methodology showing the balance between the type of the sensor sensor types and the model
capabilities in order to obtain a unique resolution to the source identification problem. Simulation-
optimisation method by Guan et al. [42] was developed to solve a nonlinear CSI problem for a complex
water distribution system. The method employed the tools developed by Rossman as a simulator to
randomly compute the concentrations at designated monitoring stations. Thereafter, the obtained
information was utilised in a continuous optimal predictor–corrector algorithm in order to identify the
source of the contamination. A case of Dover township (Tom River), N.J water distribution system was
used to validate the proposed method. Results showed that the method was efficient and robust in
identifying locations and release histories of the contaminant sources. Nonetheless, better results can be
obtained if the number of monitoring station increases, which will directly increase the cost of
procurement and operation. Consequently, the authors recommended further investigations on the
optimal sensor placement and spatial distributions. The approach by De Sanctis et al. [33] identified the
set of probable source of contamination by applying sensor status from a monitoring station and hydraulic
parameter. The method integrated the particle backtracking algorithms and fixed rules in order to
determine the probable source junction that can have effects on the positive sensor information but not a
negative measurement. Based on the assumption that sensors are perfect, the method can identify all
locations and times of possible contamination sources using information about the network hydraulic and
contamination state on distributed sensor locations. A flow path analysis gives water flow paths and travel
times using each monitoring station and sampling time. The result of the proposed method efficiently
reduces the number of possible sources and true contamination source that belong to the candidate
group. Using this method is helpful for real-time application because of the computational competence.
Improvement on the better choice of sensor location will aid the capability of this method. A mixed-integer
linear programming (MILP) method to identify the source of contamination by using the accessible
discrete data from sparse manual data at restricted locations was established by van Thienen [118].
The effects of uncertainties in water management resource systems have been widely studied [119–121].
Understanding the uniqueness of hydraulic in water distribution systems is important for building robust models
and or performing vulnerability assessments and reasonable uncertainties that produce high degrees of
uncertainty in estimated exposure levels which was discussed by Torres et al. [122]. Thus, parameter
uncertainty is a source of error that may create a large disparity in water distribution network, particularly in the
water demand which can fluctuate widely and is unpredictable [35]. These uncertainties may be computed from
the random feature of resource conditions and natural process errors in estimated model parameters or
imprecision or fuzziness human-induced [121]. These approaches provide an approximation that is always
affected by modelling uncertainty and difficult to handle by the available model calibration approaches [35]. In
real-life applications, Preis and Ostfeld [123] proposed that only a smaller quantity of the hydraulic data is
known and the single available data from the monitoring status. Thus, they presented a methodology that adds
the hydraulics ambiguity in contamination source identification. The proposed methodology relied on an earlier
model developed by the authors embedded in a statistical structure to compute the ambiguity of a
contamination source identification. Three examples of applications were used to ascertain the performance of
the proposed method through base runs and sensitivity analysis. It is noteworthy to state that demand
perturbations were not examined assuming a uniform probability between the upper and the lower values of the
anticipated demands of each of the nodes with no consumption dependencies between different nodes.
Inclusion of this may have an effect on the result. However, the main limitation of this method is the
computational intensity, which requires further investigation in order to deal with a large network scenario.
Contamination source identification under water demand uncertainty was proposed by Vankayala et al. [124]. In
this approach, a simulator designed by Rossman [47] was utilised with
Water 2018, 10, 579
stochastic and a noisy genetic algorithm (GA) as optimisers to minimise the difference between the
simulated and observed concentrations at the sensor nodes. The random demands were synthetically
created by using a Gaussian model and AR model. In order to address the demand inconsistency in the
source identification problems, two alternatives of generic algorithm were scrutinised. Overall, four
diverse scenarios were inspected on the sample network using two demand models and two optimisation
schemes. The two optimisation schemes (Stochastic and noisy GA) identified the true source as the most
feasible source for both demand models. Remarkably, GA exhibited robustness of better precision than
the stochastic. In addition, the probability of identifying the true source was lower for the more realistic AR
model than the Gaussian model. This shows that AR model is better suited in real application as it leads
to a more conservative assessment of uncertainty in the source identification problem. More research
efforts should be devoted to the influence of temporal demand uncertainty on a large water networks and
multiple contaminant sources. The challenge of a limited number of sensors, large network nodes and
varied water demands are major contributors of uncertainties in this domain. Large amount of studies has
considered the simulation–optimisation model for the CSI problem as a unimodal optimisation problem.
However, Yan et al. [125] formulated it a multimodal optimisation problem by using dynamic niching that
calculates the multiple contamination sources through one operation that provides the possibility for
screening the true contamination source in order to address the issue of uncertainty. A standard network
example in the study by Ostfeld et al. [19] was used to test the effectiveness of the developed method.
However, the authors suggested further research on this method as problems can be wide when the
network nodes are more than 1000 and water demands that consumers make real-time changes.
Substantial effort is in progress to develop an algorithm in support of the contamination events in
water distribution networks. Of interest is the application of the least square method by Hill et al. [ 126] to
minimise the deviation between the measured and simulated concentration in an attempt to locate the
source of contamination in the water network. Several previous studies developed their formulations on
the ground that continuous sensors will offer unbiased measurements of concentration. However, based
on recent advances in technology, it is possible that Boolean type sensors will be more suitable to employ
as it will indicate the occurrence of contaminant perfectly if the concentration is higher than the threshold.
The correlation result between the constant readings and Boolean measurements are presented in this
work. Indication emerged that location can be more accurate if the high quality of Boolean values was
utilised in the least square formulation. Still, the Boolean measurement algorithm failed to calculate the
injection magnitude because of the disconnected reaction between the network delay and the magnitude
of concentrations. An optimisation method developed by Cristo and Leopardi [127] aimed to locate an
accidental contamination source in water network by using the water fraction matrix concept. The method
starts from the concentration data to select a set of candidate nodes from which the source location was
identified to minimise the variation between estimated and measured concentrations. An uncertainty
analysis was presented to exhibit the capability of the method against ambiguities in concentration
measurements and water demands. The study considered a literature network of Anytown to
demonstrate the effectiveness of the method, and the authors recommended further studies on the
methodology for deliberate contaminations and multiple pollution sources. Application of the simulation–
optimisation technique to reveal a contamination sources in a laboratory water distribution system by
employing a sodium hypochlorite solution as a contaminant was reported by Mou et al. [ 128]. Output for
various input factors was compared, and their effects of the procedure were also outlined. The work of
Tryby et al. [129] suggested a methodology as a nonlinear combinatorial optimisation problem and
resolved it by utilising a genetic algorithm. The performance of the optimized scheme was evaluated and
the results verified the significance of the monitoring sensors compared to the solution proposed to
identify the sources. Liu et al. [43,44] proposed an adaptive dynamic optimisation technique (ADOPT) that
generates a real-time reaction to a contamination incident. A new multiple population-based search that
employed an evolutionary algorithm (EA) was examined to avoid the untimely convergence of EA, which
might lead to an incorrect solution. This procedure was implemented to keep a set of alternative solutions
that signify
Water 2018, 10, 579
various non-unique solutions. As new observations are incorporated, the EA outcomes produced an
improved solution that reduces the degree of non-uniqueness. Two water networks were examined
to exhibit the performance of the proposed techniques. However, the ADOPT solved a limited set of
cases based on some assumptions like a single injection location, non-reactive contaminant, and
known demand, which may be practically impossible in real life. Further investigation of this
technique is necessary to handle cases of relatively large networks.
Gugat et al. [130] described a WDN where the contaminant intrusion could occur at a limited number
of nodes. The method was based on the data collected in a defined set of sensors. The corresponding
infinite-dimensional optimisation problem was defined in a Hilbert space setting with the addition of a
quadratic regularisation term added in the objective function to guarantee a unique solution. Under
certain assumptions, the computation of the solution on a discrete time grid was performed by solving
finite-dimensional linear least squares problems. This method was considered useful to minimise
potential impacts of contamination emergencies on consumers by helping to select locations to flush the
contaminant out of the distribution network. Recently, the study by Xu et al. [131] formulated a cultural
algorithm to solve the CSI problem. Three different sizes of water supply networks were used to examine
the efficiency of the method. The results showed that the algorithm is feasible and demonstrated
robustness of the proposed method. However, excessive computational time was observed to process
the results that were a difficulty of this application. The problems associated with the contamination
source identification concern include a limited number of sensors, a large number of network nodes and
dynamic water demand that creates uncertainties. In the work of Yan et al. [132], an optimisation problem
was established and the hybrid encoding method was employed to code the problem, according to the
properties of a variable so as to improve the convergence speed and accuracy. Different sizes of the pipe
network data validated the proposed methodology. The study by Xue et al. [ 133] addressed the challenge
of water demand with the aid of a Gaussian model and then proposed an improved generic algorithm to
solve the CSI problem under uncertainty of water demand. The proposed method was experimented on
two water distribution networks of different sizes. However, they established that constant changes of
water demand and large scale network nodes can be considered as an abstract optimisation problem that
is dynamic and a multimodal function in nature. Therefore, the authors advised the need to further
investigations on an improved method to address the challenge of dynamic, large scale and multi-model
problems.
An adjoint method is a means for direct computation of the sensitivity of a system state at a specific
position and time to the dynamic system. This approach is referred to as adjoint state and a state variable
of the forward contaminant transport equations. In the case of CSI, the system state is the contaminant
transport of a monitoring station and the system parameter is the extent of discharge from the source.
The method has been utilised for source identification in groundwater [134]. The application of this
method in water distribution networks has been restricted to moderately small networks [135]. Locating of
a contamination source in a water drinking system by using this method was introduced by Neupauer
[135]. They employed the simulator approach by Rossman [47] as a tool that assumes an idea mixing at
all nodes. However, the investigation by Austin et al. [136] revealed that the proposition generated an
invalid result. This led to the development of an adjoint theory by Wang et al. [ 137], which incorporates
incomplete mixing at the junction. A coupled adjoint scheme and a Monte Carlo sensor installment
procedure was established by Ung et al. [138]. At first, the backtracking algorithm applied binary response
to compute the possible location of contamination characteristic. After which, a definite condition was
applied to precision and explicitness. Subsequently, two optimising procedures that maximise the set
conditions were suggested: greedy and a local search procedure that was integrated with a Monte Carlo
scheme to assign the best location of monitoring stations suitable for assigning the source of a
contamination. This method was tested on a practical French network that has over 2500 pipes to validate
the proposed method. The authors agreed that the choice of sensor placement for source identification
depends on the network properties along a
Water 2018, 10, 579
number of available sensors. They advocated further study on the impact of hydraulic and transport
model errors in the final output design.
to address the contamination source identification problem was discussed by Wang and Zhou [156].
Validation of this method was examined on the real-life simple network scenario. The results verified
the effectiveness of the method. One notable contribution of this method is that the probability
distribution of likely observations is organised in ranked tree and combinatorial disruption is avoided.
Additionally, a variation analysis of posterity was conducted to provide significance probability to the
achieved identification output, and the computational load in the method increases exponentially as
the number of contamination parameters increases, which limits the application of the proposed
method in the multi-parameter identification case. Consequently, studies on minimising the
computational load in order to extend the proposed method needed to be done. Another interesting
area that requires further study is the optimisation of sensor layout. The study by Baradouzi et al.
[157] based their method on Probabilistic Support Vector Machines (PSVMs) for identifying the
contamination source in water distribution systems. Several simulations were performed in order to
obtain the necessary information for training the PSVMs. The trained probabilistic SVMs were found
to be productive in locating the upstream areas that are liable to have positive identification output.
The efficiency and usefulness of the proposed approach were validated using the available
information acquired from a water distribution network of Arak in the western part of Iran.
sensor outputs leading to detection and identification of all simulated contamination events. Then, the
relative changes calculated from adaptively transformed residual chlorine measurements were
quantitatively related to contaminant-chlorine reactivity in drinking water. The results showed that the
tested contaminants were distinguishable based on kinetic and chemical differences. The approach by
Zechman and Ranjithan [159] described an approach based on evolutionary techniques and a
population-based global search procedure. The method was formulated by using a tree-based encoding
model that presented the decision vectors and a group of related genetic operators that lead to an
effective searching. The results of the technique indicated a good correlation and its capability on the
numbers of cases examined. The combination of data from several monitoring sensors to minimise false
background alarms was proposed by Koch et al. [166]. They used the Kulldorffs scan test to statistically
locate the important clusters of identification based on the location and time of isolated detections as
points leading to a random space-time point process. The results showed that the scan test can detect
significant clusters of events, reducing the occurrence of false alarms caused by background noise by
three orders of magnitude using the scan test. The clusters can also help to characterise the
contamination. The study by Di Nardo et al. [158] experimented with a basic backflow incident with the
introduction of cyanide into the potable water system. The location of the major entry portion of the
contaminant occurrence was indicated. Computation of the lethal dose of the cyanide consumed by the
consumers and the length of the affected pipes was examined along the sub-division of the isolated
region. The output of the techniques showed that sectorisation of WDN may minimise the propagation of
polluted water and prevent the customers from using unsafe water. Consequently, exposure to the
contaminated water has significant negative effects on the consumers. Hence, the author concluded
additional examination on the proposal of a water district in water networks for adequate protection. A
computational approach based on decision trees for selecting a series of junctions in the WDN to conduct
an extended evaluation of the effect of water contamination and isolating the source area with a few
quality samples was described by Eliades and Polycarpou [167]. The performance of the procedure was
demonstrated on a simplified benchmark water distribution network for validation.
Liu et al. [162] presented a methodology that integrates a statistical procedure and a heuristic
search model to characterise the contamination occurrence. The statistical procedure was employed to
spot the probable locations for the contaminant intrusion and the heuristic search method enhanced
refinement of the contamination source characteristics. Two illustrative examples of WDNs demonstrated
the ability of the method to adaptively discover the contaminant source characteristics as well as
assessing the degree of non-uniqueness. The method can be extended to solve larger networks such as
multi-source contaminant. However, the scenarios examined here did not account for uncertainty of the
hydraulic model, which has implications on the output. Hence, future research should be conducted to
cater for the uncertainty that can lead to the excessive computational burden that can pose a challenge
on both accuracy and efficiency of the method. The effort made by Liu et al. [161] introduced a hybrid
method for the real-time characterisation of a contaminant source given sensor measurements in WDNs.
This approach incorporates a logistic regression (LR) and local improvement model to accelerate the
convergence. The study by Liu et al. [43] employed an evolutionary strategy (ES)- based Adaptive
Dynamic Optimisation Technique (ADOPT) to adaptively examine the group of contaminant source
characteristics. Actually, the ADOPT exhibits a promising result, but requires an extensive computational
to converge. This computational burden will not only affect the identification time, but also the quality of
the result. In order to minimise the simulation time and sustain the adequacy of the accuracy, the ADOPT
techniques require further improvement. Here, the integration of LS into the ADOPT stimulates the
performance of the algorithm of the ADOPT, and the degree of improvements may depend on the
problem. This augmentation assists the operation of ADOPT by eradicating unnecessary nodes and
adjusting the process. To exhibit the efficiency of the hybrid model, two networks were considered. One is
a small network of about 117 pipes and the other is a relatively large network with details in [168]. Several
interesting results were derived from the networks examined. The hybrid methodology has a faster
convergence than a standalone approach.
Water 2018, 10, 579
This is evidence that the hybrid process can improve not only on the quality of the solution, but also
on the computational efficiency, particularly when a larger network is considered. The procedure of
local search selection plays a vital role in the algorithm performance, which requires further study,
especially under various degrees of complex scenarios. Costa et al. [169] investigated the challenge
of restricting the contamination occurrence by using information collected from the monitoring
stations. The proposed approach was tested on a real-life network to verify the methodology. The
outcome of the investigation revealed that it is possible to restrict the sources of contamination to
the first sensor that detects it. Thereafter, an improvement on the outcome can be made as new
information is accessible. This indicates the possibility of detecting the contaminant concentration
variation in the network at the new sensor location. Thus, the other potential contamination event
can be identified and location of sources can be further confined. Two scenarios were also
examined to ascertain the effect of irregularities of false positives. The results demonstrated that
false positive incident has no connection to the real identification.
10. Conclusions
Identification of contamination source in a water distribution network is essential considering the
consequences the use of contaminated water on the public. In this study, a comprehensive survey on the
existing contamination source identification methods is presented. From this survey, it is evident that the
existing approaches were able to identify the contamination source to a reasonable extent. However,
each of the methods has certain limitations as reported in the literature. Associated technical challenges
include uncertainties of sensors, stochastic water demands, excessive computational time, particularly for
a large network among others. Despite a series of investigations on the contamination source
identification problem, the research in the field is still at an infancy stage. Hence, further investigation is
essential in order to properly address the research gaps. Generally, contamination source identification
remains a topical issue in a water distribution network due to the complex nature of the system and its
significance in the economy and social instability. More research efforts should be devoted to the
applicability of these techniques to ascertain their efficiency in a real-life scenario in order to minimise the
economic and public health consequences the use of contaminated water.
References
1. Swamee, P.K.; Sharma, A.K. Design of Water Supply Pipe Networks; John Wiley & Sons: Hoboken, NJ, USA, 2008.
2. Yazdani, A.; Otoo, R.A.; Jeffrey, P. Resilience enhancing expansion strategies for water distribution
systems: A network theory approach. Environ. Model. Softw. 2011, 26, 1574–1582. [CrossRef]
3. Collins, M.; Cooper, L.; Helgason, R.; Kennington, J.; LeBlanc, L. Solving the pipe network analysis
problem using optimization techniques. Manag. Sci. 1978, 24, 747–760. [CrossRef]
4. Clark, R.M.; Grayman, W.M.; Males, R.M.; Hess, A.F. Modeling contaminant propagation in drinking-
water distribution systems. J. Environ. Eng. 1993, 119, 349–364. [CrossRef]
5. Clark, R.M.; Grayman, W.M.; Goodrich, J.A.; Deininger, R.A.; Skov, K. Measuring and modeling chlorine
propagation in water distribution systems. J. Water Resour. Plan. Manag. 1994, 120, 871–887. [CrossRef]
6. Clark, R.M.; Grayman, W.M.; Males, R.M. Contaminant propagation in distribution systems. J. Environ.
Eng. 1988, 114, 929–943. [CrossRef]
7. Clark, R.M.; Deininger, R.A. Protecting the nation’s critical infrastructure: The vulnerability of US water
supply systems. J. Conting. Crisis Manag. 2000, 8, 73–80. [CrossRef]
8. Kirmeyer, G.J.; Martel, K. Pathogen Intrusion into the Distribution System; American Water Works
Association: Denver, CO, USA, 2001.
9. Mac Kenzie, W.R.; Hoxie, N.J.; Proctor, M.E.; Gradus, M.S.; Blair, K.A.; Peterson, D.E.; Kazmierczak, J.J.;
Addiss, D.G.; Fox, K.R.; Rose, J.B.; et al. A massive outbreak in Milwaukee of Cryptosporidium infection
transmitted through the public water supply. N. Engl. J. Med. 1994, 331, 161–167. [CrossRef] [PubMed]
10. Corso, P.S.; Kramer, M.H.; Blair, K.A.; Addiss, D.G.; Davis, J.P.; Haddix, A.C. Costs of illness in the 1993
waterborne Cryptosporidium outbreak, Milwaukee, Wisconsin. Emerg. Infect. Dis. 2003, 9, 426.
[CrossRef] [PubMed]
11. Cooper, W.J. Responding to crisis: The West Virginia chemical spill. Environ. Sci. Technol. 2014, 48,
3095–3095. [CrossRef] [PubMed]
12. Berry, J.W.; Fleischer, L.; Hart, W.E.; Phillips, C.A.; Watson, J.P. Sensor placement in municipal water
networks. J. Water Resour. Plan. Manag. 2005, 131, 237–243. [CrossRef]
13. Berry, J.W.; Hart, W.E.; Phillips, C.A.; Uber, J.G.; Watson, J.P. Validation and assessment of integer
programming sensor placement models. In Impacts of Global Climate Change; American Society of Civil
Engineers: Reston, VA, USA, 2005; pp. 1–11.
Water 2018, 10, 579
14. Berry, J.; Hart, W.E.; Phillips, C.A.; Uber, J.G.; Watson, J.P. Sensor placement in municipal water networks with
temporal integer programming models. J. Water Resour. Plan. Manag. 2006, 132, 218–224. [CrossRef]
15. Berry, J.; Carr, R.D.; Hart, W.E.; Leung, V.J.; Phillips, C.A.; Watson, J.P. Designing contamination
warning systems for municipal water networks using imperfect sensors. J. Water Resour. Plan. Manag.
2009, 135, 253–263. [CrossRef]
16. Ostfeld, A.; Salomons, E. Optimal layout of early warning detection stations for water distribution systems
security. J. Water Resour. Plan. Manag. 2004, 130, 377–385. [CrossRef]
17. Ostfeld, A.; Salomons, E. Optimal early warning monitoring system layout for water networks security: Inclusion
of sensors sensitivities and response delays. Civ. Eng. Environ. Syst. 2005, 22, 151–169. [CrossRef]
18. Ostfeld, A.; Salomons, E. Securing water distribution systems using online contamination monitoring. J.
Water Resour. Plan. Manag. 2005, 131, 402–405. [CrossRef]
19. Ostfeld, A.; Uber, J.G.; Salomons, E.; Berry, J.W.; Hart, W.E.; Phillips, C.A.; Watson, J.P.; Dorini, G.;
Jonkergouw, P.; Kapelan, Z.; et al. The battle of the water sensor networks (BWSN): A design challenge
for engineers and algorithms. J. Water Resour. Plan. Manag. 2008, 134, 556–568. [CrossRef]
20. Propato, M. Contamination warning in water networks: General mixed-integer linear models for sensor
location design. J. Water Resour. Plan. Manag. 2006, 132, 225–233. [CrossRef]
21. Kansal, M.; Dorji, T.; Chandniha, S.K. Design scheme for water quality monitoring in a distribution
network. Int. J. Environ. Dev. 2012, 9, 69–81.
22. Krause, A.; Leskovec, J.; Guestrin, C.; VanBriesen, J.; Faloutsos, C. Efficient sensor placement optimization for
securing large water distribution networks. J. Water Resour. Plan. Manag. 2008, 134, 516–526. [CrossRef]
23. Afshar, A.; Mariño, M.A. Multi-objective coverage-based ACO model for quality monitoring in large water
networks. Water Resour. Manag. 2012, 26, 2159–2176. [CrossRef]
24. Afshar, A.; Khombi, S.M. Multiobjective Optimization of Sensor Placement in Water Distribution Networks;
Dual Use Benefit Approach. Int. J. Optim. Civ. Eng. 2015, 5, 315–331.
25. Woo, H.M.; Yoon, J.H.; Choi, D.Y. Optimal monitoring sites based on water quality and quantity in water
distribution systems. In Bridging the Gap: Meeting the World’s Water and Environmental Resources
Challenges; American Society of Civil Engineers: Reston, VA, USA, 2001; pp. 1–9.
26. Cozzolino, L.; Mucherino, C.; Pianese, D.; Pirozzi, F. Positioning, within water distribution networks, of
monitoring stations aiming at an early detection of intentional contamination. Civ. Eng. Environ. Syst.
2006, 23, 161–174. [CrossRef]
27. Adedeji, K.B.; Hamam, Y.; Abe, B.T.; Abu-Mahfouz, A.M. Leakage Detection and Estimation Algorithm for
Loss Reduction in Water Piping Networks. Water 2017, 9, 773. [CrossRef]
28. Hindi, K.; Hamam, Y. Locating pressure control elements for leakage minimization in water supply
networks: An optimization model. Eng. Optim. 1991, 17, 281–291. [CrossRef]
29. Hindi, K.; Hamam, Y. Pressure control for leakage minimization in water supply networks Part 1: Single
period models. Int. J. Syst. Sci. 1991, 22, 1573–1585. [CrossRef]
30. Hindi, K.; Hamam, Y. Locating pressure control elements for leakage minimisation in water supply
networks by genetic algorithms. In Artificial Neural Nets and Genetic Algorithms; Springer: Vienna,
Austria, 1993; pp. 583–587.
31. Sun, N.Z. Inverse Problems in Groundwater Modeling; Springer Science & Business Media: Berlin,
Germany, 2013; Volume 6.
32. Laird, C.D.; Biegler, L.T.; van Bloemen Waanders, B.G.; Bartlett, R.A. Contamination source
determination for water networks. J. Water Resour. Plan. Manag. 2005, 131, 125–134. [CrossRef]
33. De Sanctis, A.E.; Shang, F.; Uber, J.G. Real-time identification of possible contamination sources using
network backtracking methods. J. Water Resour. Plan. Manag. 2009, 136, 444–453. [CrossRef]
34. Wang, H.; Harrison, K.W. Improving efficiency of the Bayesian approach to water distribution contaminant
source characterization with support vector regression. J. Water Resour. Plan. Manag. 2012, 140, 3–11.
[CrossRef]
35. Huang, J.J.; McBean, E.A.; James, W. Multi-objective optimization for monitoring sensor placement in
water distribution systems. In Water Distribution Systems Analysis Symposium 2006; American Society
of Civil Engineers: Reston, VA, USA, 2008; pp. 1–14.
Water 2018, 10, 579
36. Tsitsifli, S.; Kanakoudis, V. Drinking water quality and safety assessment—A review. In Proceedings of
the 6th International Conference on Environmental Management, Planning, Engineering (CEMEPE2017),
Thessaloniki, Greece, 25–30 June 2017; pp. 403–416.
37. Kanakoudis, V.; Tsitsifli, S. Potable water security assessment—A review on monitoring, modelling and
optimization techniques, applied to water distribution networks. Desalin. Water Treat. 2017, 99, 18–26.
[CrossRef]
38. Deng, W.; Wang, G. A novel water quality data analysis framework based on time-series data mining.
J. Environ. Manag. 2017, 196, 365–375. [CrossRef] [PubMed]
39. Hu, C.; Zhao, J.; Yan, X.; Zeng, D.; Guo, S. A MapReduce based parallel niche genetic algorithm for
contaminant source identification in water distribution network. Ad Hoc Netw. 2015, 35, 116–126. [CrossRef]
40. Perelman, L.; Ostfeld, A. Bayesian networks for source intrusion detection. J. Water Resour. Plan.
Manag. 2012, 139, 426–432. [CrossRef]
41. Wagner, D.; Neupauer, R. Probabilistic contaminant source identification in water distribution systems
with incomplete mixing at pipe junctions. In World Environmental and Water Resources Congress 2013:
Showcasing the Future; American Society of Civil Engineers: Reston, VA, USA, 2013; pp. 930–935.
42. Guan, J.; Aral, M.M.; Maslia, M.L.; Grayman, W.M. Identification of contaminant sources in water
distribution systems using simulation-optimization method: Case study. J. Water Resour. Plan. Manag.
2006, 132, 252–262. [CrossRef]
43. Liu, L.; Ranjithan, S.R.; Mahinthakumar, G. Contamination source identification in water distribution
systems using an adaptive dynamic optimization procedure. J. Water Resour. Plan. Manag. 2010, 137,
183–192. [CrossRef]
44. Liu, L.; Zechman, E.M.; Brill, E.D., Jr.; Mahinthakumar, G.; Ranjithan, S.; Uber, J. Adaptive contamination
source identification in water distribution systems using an evolutionary algorithm-based dynamic
optimization procedure. In Water Distribution Systems Analysis Symposium 2006; American Society of
Civil Engineers: Reston, VA, USA, 2008; pp. 1–9.
45. Shang, F.; Uber, J.; Murray, R. Uncertainty of contamination source characterization within water
distribution systems. In World Environmental and Water Resources Congress 2007: Restoring Our
Natural Habitat; American Society of Civil Engineers: Reston, VA, USA, 2007; pp. 1–8.
46. Murray, R.; Haxton, T.; Janke, R.; Hart, W.E.; Berry, J.; Phillips, C. Sensor Network Design for Drinking
Water Contamination Warning Systems: A Compendium of Research Results and Case Studies Using
the TEVA-SPOT Software; National Homeland Security Research Center, US Environmental Protection
Agency: Cincinnati, OH, USA, 2010.
47. Rossman, L.A. EPANET 2: Users Manual; National Risk Management Research Laboratory: Cincinnati,
OH, USA, 2000.
48. Tinelli, S.; Creaco, E.; Ciaponi, C. Sampling significant contamination events for optimal sensor
placement in water distribution systems. J. Water Resour. Plan. Manag. 2017, 143, 04017058. [CrossRef]
49. Davidson, J.; Bouchart, F.; Cavill, S.; Jowitt, P. Real-time connectivity modeling of water distribution
networks to predict contamination spread. J. Comput. Civ. Eng. 2005, 19, 377–386. [CrossRef]
50. Boulus, P.F.; Altman, T.; Sadhal, K.S. Computer modeling of water quality in large multiple-source
networks. Appl. Math. Model. 1992, 16, 439–445. [CrossRef]
51. Males, R.M.; Clark, R.M.; Wehrman, P.J.; Gates, W.E. Algorithm for mixing problems in water systems.
J. Hydraul. Eng. 1985, 111, 206–219. [CrossRef]
52. Wood, D.J.; Ormsbee, L.E. Supply identification for water distribution systems. J. Am. Water Works
Assoc. 1989, 81, 74–80. [CrossRef]
53. Bhave, P.R. Extended period simulation of water systems—Direct solution. J. Environ. Eng. 1988, 114,
1146–1159. [CrossRef]
54. Creaco, E.; Pezzinga, G. Embedding linear programming in multi objective genetic algorithms for reducing
the size of the search space with application to leakage minimization in water distribution networks.
Environ. Model. Softw. 2015, 69, 308–318. [CrossRef]
55. Giustolisi, O.; Kapelan, Z.; Savic, D. Extended period simulation analysis considering valve shutdowns.
J. Water Resour. Plan. Manag. 2008, 134, 527–537. [CrossRef]
56. Giustolisi, O.; Laucelli, D.; Berardi, L.; Savi´c, D.A. Computationally efficient modeling method for large
water network analysis. J. Hydraul. Eng. 2011, 138, 313–326. [CrossRef]
Water 2018, 10, 579
57. Todini, E. A more realistic approach to the extended period simulation of water distribution networks. Adv.
Water Supply Manag. 2003, 173–184.
58. Rossman, L.A.; Boulos, P.F. Numerical methods for modeling water quality in distribution systems: A
comparison. J. Water Resour. Plan. Manag. 1996, 122, 137–146. [CrossRef]
59. Liou, C.; Kroon, J. Modeling the propagation of waterborne substances in distribution networks. J. Am.
Water Works Assoc. 1987, 79, 54–58. [CrossRef]
60. Boulos, P.F.; Altman, T.; Jarrige, P.A.; Collevati, F. Discrete simulation approach for network-water-
quality models. J. Water Resour. Plan. Manag. 1995, 121, 49–60. [CrossRef]
61. Boulos, P.F.; Altman, T.; Jarrige, P.A.; Collevati, F. An event-driven method for modelling contaminant
propagation in water networks. Appl. Math. Model. 1994, 18, 84–92. [CrossRef]
62. Islam, M.R.; Chaudhry, M.H. Modeling of constituent transport in unsteady flows in pipe networks.
J. Hydraul. Eng. 1998, 124, 1115–1124. [CrossRef]
63. Grayman, W.M.; Clark, R.M.; Males, R.M. Modeling distribution-system water quality; dynamic approach.
J. Water Resour. Plan. Manag. 1988, 114, 295–312. [CrossRef]
64. Rossman, L.A.; Boulos, P.F.; Altman, T. Discrete volume-element method for network water-quality models.
J. Water Resour. Plan. Manag. 1993, 119, 505–517. [CrossRef]
65. Rossman, L.A.; Clark, R.M.; Grayman, W.M. Modeling chlorine residuals in drinking-water distribution
systems. J. Environ. Eng. 1994, 120, 803–820. [CrossRef]
66. Mansour-Rezaei, S.; Naser, G.; Malekpour, A.; Karney, B.W. Contaminant intrusion in water distribution
systems. J. Am. Water Works Assoc. 2013, 105, E278–E290. [CrossRef]
67. Fernandes, C.; Karney, B. Modelling the advection equation under water hammer conditions. Urban
Water J. 2004, 1, 97–112. [CrossRef]
68. Tzatchkov, V.G.; Aldama, A.A.; Arreguin, F.I. Advection-dispersion-reaction modeling in water distribution
networks. J. Water Resour. Plan. Manag. 2002, 128, 334–342. [CrossRef]
69. Basha, H.; Malaeb, L. Eulerian-Lagrangian method for constituent transport in water distribution networks.
J. Hydraul. Eng. 2007, 133, 1155–1166. [CrossRef]
70. Munavalli, G.; Kumar, M.M. Modified Lagrangian method for modeling water quality in distribution
systems. Water Res. 2004, 38, 2973–2988. [CrossRef] [PubMed]
71. Cross, H. Analysis of Flow in Networks of Conduits or Conductors; Technical Report; University of Illinois at
Urbana Champaign, College of Engineering, Engineering Experiment Station: Champaign, IL, USA, 1936.
72. Martin, D.; Peters, G. The application of Newtons method to network analysis by digital computer. J. Inst.
Water Eng. 1963, 17, 115–129.
73. Epp, R.; Fowler, A.G. Efficient code for steady-state flows in networks. J. Hydraul. Div. 1970, 96, 43–56.
74. Hamam, Y.; Brameller, A. Hybrid method for the solution of piping networks. Proc. Inst. Electr. Eng. 1971,
118, 1607–1612. [CrossRef]
75. Wood, D.J.; Charles, C.O. Hydraulic network analysis using linear theory. J. Hydraul. Div. 1972, 98, 1157–1170.
76. Todini, E.; Pilati, S. A gradient method for the analysis of pipe networks. In Proceedings of the
International Conference on Computer Applications for Water Supply and Distribution, Leicester
Polytechnic, UK, 8–10 September 1987.
77. Todini, E. A unifying view on the different looped pipe network analysis algorithms. In Computing and
Control for the Water Industry; Research Studies Press Ltd.: Baldock, UK, 1999; pp. 63–80.
78. Todini, E.; Rossman, L.A. Unified framework for deriving simultaneous equation algorithms for water
distribution networks. J. Hydraul. Eng. 2012, 139, 511–526. [CrossRef]
79. Walski, T.M.; Chase, D.V.; Savic, D.A.; Grayman, W.; Beckwith, S.; Koelle, E. Advanced Water
Distribution Modeling and Management. Civ. Environ. Eng. Eng. Mech. Fac. Publ. 2003, 18.
80. Strafaci, A. Advanced Water Distribution Modeling and Management. Ed; Haestad Press: Waterbury, CT,
USA, 2003; pp. 577–583.
81. Clark, R.M.; Rossman, L.A.; Wymer, L.J. Modeling distribution system water quality: Regulatory implications.
J. Water Resour. Plan. Manag. 1995, 121, 423–428. [CrossRef]
82. Ozdemir, O.N.; Ger, A.M. Unsteady 2-D chlorine transport in water supply pipes. Water Res. 1999, 33,
3637–3645. [CrossRef]
83. Al-Omari, A.S.; Chaudhry, M.H. Unsteady-state inverse chlorine modeling in pipe networks. J.
Hydraul. Eng. 2001, 127, 669–677. [CrossRef]
Water 2018, 10, 579
84. Ozdemir, O.N.; Ucak, A. Simulation of chlorine decay in drinking-water distribution systems. J. Environ.
Eng. 2002, 128, 31–39. [CrossRef]
85. Boulos, P.; Altman, T. Explicit calculation of water quality parameters in pipe distribution systems. Civ.
Eng. Syst. 1993, 10, 187–206. [CrossRef]
86. Mau, R.E.; Boulos, P.F.; Bowcock, R.W. Modelling distribution storage water quality: An analytical
approach. Appl. Math. Model. 1996, 20, 329–338. [CrossRef]
87. Islam, M.R.; Chaudhry, M.H. Numerical solution of transport equation for applications in environmental
hydraulics and hydrology. J. Hydrol. 1997, 191, 106–121. [CrossRef]
88. Islam, M.R.; Chaudhry, M.H.; Clark, R.M. Inverse modeling of chlorine concentration in pipe networks
under dynamic condition. J. Environ. Eng. 1997, 123, 1033–1040. [CrossRef]
89. Zierolf, M.L.; Polycarpou, M.M.; Uber, J.G. Development and autocalibration of an input-output model of
chlorine transport in drinking water distribution systems. IEEE Trans. Control Syst. Technol. 1998, 6, 543–
553. [CrossRef]
90. Shang, F.; Uber, J.G.; Polycarpou, M.M. Particle backtracking algorithm for water distribution system
analysis. J. Environ. Eng. 2002, 128, 441–450. [CrossRef]
91. Al-Zahrani, M.A.; Moied, K. Optimizing water quality monitoring stations using genetic algorithms. Arab. J.
Sci. Eng. 2003, 28, 57–75.
92. Ewing, R.E.; Wang, H. A summary of numerical methods for time-dependent advection-dominated partial
differential equations. In Partial Differential Equations; Elsevier: New York, NY, USA, 2001; pp. 423–445.
93. Alhumaizi, K. Comparison of finite difference methods for the numerical simulation of reacting flow.
Comput. Chem. Eng. 2004, 28, 1759–1769. [CrossRef]
94. Alhumaizi, K. Flux-limiting solution techniques for simulation of reaction–diffusion–convection system.
Commun. Nonlinear Sci. Numer. Simul. 2007, 12, 953–965. [CrossRef]
95. Hallam, N.; West, J.; Forster, C.; Powell, J.; Spencer, I. The decay of chlorine associated with the pipe
wall in water distribution systems. Water Res. 2002, 36, 3479–3488. [CrossRef]
96. Munavalli, G.; Kumar, M.M. Water quality parameter estimation in steady-state distribution system. J.
Water Resour. Plan. Manag. 2003, 129, 124–134. [CrossRef]
97. Munavalli, G.; Kumar, M.M. Water quality parameter estimation in a distribution system under dynamic
state. Water Res. 2005, 39, 4287–4298. [CrossRef] [PubMed]
98. Yang, Y.J.; Haught, R.C.; Goodrich, J.A. Real-time contaminant detection and classification in a drinking
water pipe using conventional water quality sensors: Techniques and experimental results. J. Environ.
Manag. 2009, 90, 2494–2506. [CrossRef] [PubMed]
99. Gancel, G.; Mortazavi, I.; Piller, O. Coupled numerical simulation and sensitivity assessment for quality
modelling for water distribution systems. Appl. Math. Lett. 2006, 19, 1313–1319. [CrossRef]
100. Siew, C.; Tanyimboh, T.T. Pressure-dependent EPANET extension: Extended period simulation. In Water
Distribution Systems Analysis 2010; American Society of Civil Engineers: Reston, VA, USA, 2010; pp. 85–95.
101. Jun, L.; Guoping, Y. Iterative methodology of pressure-dependent demand based on EPANET for pressure-
deficient water distribution analysis. J. Water Resour. Plan. Manag. 2012, 139, 34–44. [CrossRef]
102. Muranho, J.; Ferreira, A.; Sousa, J.; Gomes, A.; Marques, A.S. Pressure-dependent demand and leakage
modelling with an EPANET extension—WaterNetGen. Proc. Eng. 2014, 89, 632–639. [CrossRef]
103. Conejos, M.P.; Alzamora, F.M.; Alonso, J.C. A water distribution system model to simulate critical scenarios by
considering both leakage and pressure dependent demands. Proc. Eng. 2017, 186, 380–387. [CrossRef]
104. Zhang, G.Y. Real-Time Simulation of Pipe Network Based on EPANET. In Applied Mechanics and
Materials; Trans Tech Publications: Zürich, Switzerland, 2014; Volume 548, pp. 1800–1803.
105. Seyoum, A.G.; Tanyimboh, T.T. Investigation into the pressure-driven extension of the EPANET hydraulic
simulation model for water distribution systems. Water Resour. Manag. 2016, 30, 5351–5367. [CrossRef]
106. Puri, D.; Borel, K.; Vance, C.; Karthikeyan, R. Optimization of a Water Quality Monitoring Network Using a
Spatially Referenced Water Quality Model and a Genetic Algorithm. Water 2017, 9, 704. [CrossRef]
107. Huang, L.; Wang, L.; Zhang, Y.; Xing, L.; Hao, Q.; Xiao, Y.; Yang, L.; Zhu, H. Identification of
Groundwater Pollution Sources by a SCE-UA Algorithm-Based Simulation/Optimization Model. Water
2018, 10, 193. [CrossRef]
108. Huang, P.S.; Chiu, Y.C. A Simulation-Optimization Model for Seawater Intrusion Management at
Pingtung Coastal Area, Taiwan. Water 2018, 10, 251. [CrossRef]
Water 2018, 10, 579
109. Anele, A.O.; Hamam, Y.; Abu-Mahfouz, A.M.; Todini, E. Overview, Comparative Assessment and
Recommendations of Forecasting Models for Short-Term Water Demand Prediction. Water 2017, 9, 887.
[CrossRef]
110. Letting, L.K.; Hamam, Y.; Abu-Mahfouz, A.M. Estimation of Water Demand in Water Distribution Systems
Using Particle Swarm Optimization. Water 2017, 9, 593. [CrossRef]
111. Lee, Y.J.; Park, C.; Lee, M.L. Identification of a Contaminant Source Location in a River System Using
Random Forest Models. Water 2018, 10, 391. [CrossRef]
112. van Bloemen Waanders, B.G.; Bartlett, R.A.; Biegler, L.T.; Laird, C.D. Nonlinear programming strategies
for source detection of municipal water networks. In World Water & Environmental Resources Congress
2003; American Society of Civil Engineers: Reston, VA, USA, 2003; pp. 1–10.
113. Laird, C.D.; Biegler, L.T.; van Bloemen Waanders, B.G. Real-time, large-scale optimization of water
network systems using a subdomain approach. In Real-Time PDE-Constrained Optimization; Society for
Industrial and Applied Mathematics: Philadelphia, PA, USA, 2007; pp. 289–306.
114. Preis, A.; Ostfeld, A. A contamination source identification model for water distribution system security.
Eng. Optim. 2007, 39, 941–947. [CrossRef]
115. Preis, A.; Ostfeld, A. Multiobjective sensor design for water distribution systems security. In Water
Distribution Systems Analysis Symposium 2006; American Society of Civil Engineers: Reston, VA, USA,
2008; pp. 1–17.
116. Preis, A.; Ostfeld, A. Genetic algorithm for contaminant source characterization using imperfect sensors.
Civ. Eng. Environ. Syst. 2008, 25, 29–39. [CrossRef]
117. Prasad, T.D.; Park, N.S. Multiobjective genetic algorithms for design of water distribution networks. J.
Water Resour. Plan. Manag. 2004, 130, 73–82. [CrossRef]
118. Van Thienen, P. Alternative strategies for optimal water quality sensor placement in drinking water
distribution networks. In Proceedings of the 11th International Conference on Hydroinformatics, New
York, NY, USA, 16–21 August 2014.
119. Chung, G.; Lansey, K.; Bayraksan, G. Reliable water supply system design under uncertainty. Environ.
Model. Softw. 2009, 24, 449–462. [CrossRef]
120. Li, Y.; Huang, G.; Nie, S. Planning water resources management systems using a fuzzy-boundary
interval-stochastic programming method. Adv. Water Resour. 2010, 33, 1105–1117. [CrossRef]
121. Li, Y.; Huang, G.; Nie, S.; Chen, X. A robust modeling approach for regional water management under
multiple uncertainties. Agric. Water Manag. 2011, 98, 1577–1588. [CrossRef]
122. Torres, J.M.; Brumbelow, K.; Guikema, S.D. Risk classification and uncertainty propagation for virtual
water distribution systems. Reliab. Eng. Syst. Saf. 2009, 94, 1259–1273. [CrossRef]
123. Preis, A.; Ostfeld, A. Hydraulic uncertainty inclusion in water distribution systems contamination source
identification. Urban Water J. 2011, 8, 267–277. [CrossRef]
124. Vankayala, P.; Sankarasubramanian, A.; Ranjithan, S.R.; Mahinthakumar, G. Contaminant source
identification in water distribution networks under conditions of demand uncertainty. Environ. Forensics
2009, 10, 253–263. [CrossRef]
125. Yan, X.; Zhao, J.; Hu, C. Research on Multimodal Optimization Algorithm for the Contamination Source
Identification of City Water Distribution Networks. In Bio-Inspired Computing-Theories and Applications;
Springer: Singapore, 2016; pp. 80–85.
126. Hill, J.; van Bloemen Waanders, B.; Laird, C. Source inversion with uncertain sensor measurements. In
Proceedings of the Water Distribution System Analysis Symposium 2006, Cincinnati, OH, USA, 27–30
August 2006.
127. Cristo, C.D.; Leopardi, A. Pollution source identification of accidental contamination in water distribution
networks. J. Water Resour. Plan. Manag. 2008, 134, 197–202. [CrossRef]
128. Mou, L.; Menglin, W.; Jie, L.; Shen, D. Notice of Retraction Investigation on backward tracking of
contamination sources in water supply systems—Case study. In Proceedings of the 2010 The 2nd
Conference on Environmental Science and Information Application Technology, Wuhan, China, 17–18
July 2010; Volume 3, pp. 484–487.
129. Tryby, M.E.; Propato, M.; Ranjithan, S.R. Monitoring design for source identification in water distribution
systems. J. Water Resour. Plan. Manag. 2010, 136, 637–646. [CrossRef]
Water 2018, 10, 579
130. Gugat, M. Contamination source determination in water distribution networks. SIAM J. Appl. Math. 2012,
72, 1772–1791. [CrossRef]
131. Yan, X.; Gong, W.; Wu, Q. Contaminant source identification of water distribution networks using cultural
algorithm. Concurr. Comput. Pract. Exp. 2017, 29. [CrossRef]
132. Yan, X.; Zhao, J.; Hu, C.; Wu, Q. Contaminant source identification in water distribution network based on
hybrid encoding. J. Comput. Methods Sci. Eng. 2016, 16, 379–390. [CrossRef]
133. Xuesong, Y.; Jie, S.; Chengyu, H. Research on contaminant sources identification of uncertainty water
demand using genetic algorithm. Cluster Comput. 2017, 20, 1007–1016. [CrossRef]
134. Neupauer, R.M.; Wilson, J.L. Adjoint method for obtaining backward-in-time location and travel time
probabilities of a conservative groundwater contaminant. Water Resour. Res. 1999, 35, 3389–3398. [CrossRef]
135. Neupauer, R.M. Adjoint sensitivity analysis of contaminant concentrations in water distribution systems.
J. Eng. Mech. 2010, 137, 31–39. [CrossRef]
136. Austin, R.; Waanders, B.v.B.; McKenna, S.; Choi, C. Mixing at cross junctions in water distribution
systems. II: Experimental study. J. Water Resour. Plan. Manag. 2008, 134, 295–302. [CrossRef]
137. Wang, H.; Jin, X. Characterization of groundwater contaminant source using Bayesian method. Stoch.
Environ. Res. Risk Assess. 2013, 27, 867–876. [CrossRef]
138. Ung, H.; Piller, O.; Gilbert, D.; Mortazavi, I. Accurate and Optimal Sensor Placement for Source Identification of
Water Distribution Networks. J. Water Resour. Plan. Manag. 2017, 143, 04017032. [CrossRef]
139. Hobbs, B.F. Bayesian methods for analysing climate change and water resource uncertainties. J. Environ.
Manag. 1997, 49, 53–72. [CrossRef]
140. Bates, B.C.; Campbell, E.P. A Markov chain Monte Carlo scheme for parameter estimation and inference
in conceptual rainfall-runoff modeling. Water Resour. Res. 2001, 37, 937–947. [CrossRef]
141. Marshall, L.; Nott, D.; Sharma, A. A comparative study of Markov chain Monte Carlo methods for
conceptual rainfall-runoff modeling. Water Resour. Res. 2004, 40. [CrossRef]
142. Boccelli, D.; Uber, J. Incorporating spatial correlation in a Markov chain Monte Carlo approach for
network model calibration. Impacts Glob. Clim. Chang. 2005, 1–8. [CrossRef]
143. Soldevila, A.; Fernandez-Canti, R.M.; Blesa, J.; Tornil-Sin, S.; Puig, V. Leak localization in water
distribution networks using Bayesian classifiers. J. Process Control 2017, 55, 1–9. [CrossRef]
144. Gargano, R.; Tricarico, C.; Granata, F.; Santopietro, S.; de Marinis, G. Probabilistic models for the peak
residential water demand. Water 2017, 9, 417. [CrossRef]
145. Candelieri, A. Clustering and Support Vector Regression for Water Demand Forecasting and Anomaly
Detection. Water 2017, 9, 224. [CrossRef]
146. Shabani, S.; Candelieri, A.; Archetti, F.; Naser, G. Gene Expression Programming Coupled with
Unsupervised Learning: A Two-Stage Learning Process in Multi-Scale, Short-Term Water Demand
Forecasts. Water 2018, 10, 142. [CrossRef]
147. Candelieri, A.; Perego, R.; Archetti, F. Bayesian optimization of pump operations in water distribution
systems. J. Glob. Optim. 2018, 1–23. [CrossRef]
148. Zhang, J.; Zeng, L.; Chen, C.; Chen, D.; Wu, L. Efficient Bayesian experimental design for contaminant
source identification. Water Resour. Res. 2015, 51, 576–598. [CrossRef]
149. Dawsey, W.J.; Minsker, B.S.; VanBlaricum, V.L. Bayesian belief networks to integrate monitoring evidence of
water distribution system contamination. J. Water Resour. Plan. Manag. 2006, 132, 234–241. [CrossRef]
150. Dawsey, W.J. Bayesian Belief Networks to Integrate Monitoring Evidence of Water Distribution System
Contamination. Ph.D. Thesis, University of Illinois at Urbana-Champaign, Champaign, IL, USA, 2012.
151. Tao, T.; Huang, H.D.; Xin, K.L.; Liu, S.M. Identification of contamination source in water distribution network
based on consumer complaints. J. Central South Univ. Technol. 2012, 19, 1600–1609. [CrossRef]
152. De Sanctis, A.; Boccelli, D.; Shang, F.; Uber, J. Probabilistic approach to characterize contamination
sources with imperfect sensors. In Proceedings of the World Environmental and Water Resources
Congress 2008: Ahupua’A, Honolulu, HI, USA, 12–13 May 2008; pp. 1–10.
153. Yang, X.; Boccelli, D.L. Bayesian approach for real-time probabilistic contamination source identification.
J. Water Resour. Plan. Manag. 2013, 140, 04014019. [CrossRef]
154. Perelman, L.; Ostfeld, A. Bayesian networks for estimating contaminant source and propagation in a
water distribution system using cluster structure. In Water Distribution Systems Analysis 2010;
American Society of Civil Engineers: Reston, VA, USA, 2010; pp. 426–435.
Water 2018, 10, 579
155. Neupauer, R.M.; Records, M.K.; Ashwood, W.H. Backward probabilistic modeling to identify contaminant
sources in water distribution systems. J. Water Resour. Plan. Manag. 2009, 136, 587–591. [CrossRef]
156. Wang, C.; Zhou, S. Contamination source identification based on sequential Bayesian approach for water
distribution network with stochastic demands. IISE Trans. 2017, 49, 899–910. [CrossRef]
157. Barandouzi, M.; Kerachian, R. Probabilistic Contaminant Source Identification in Water Distribution
Infrastructure Systems. Civ. Eng. Infrastruct. J. 2016, 49, 311–326.
158. Di Nardo, A.; Di Natale, M.; Guida, M.; Musmarra, D. Water network protection from intentional
contamination by sectorization. Water Resour. Manag. 2013, 27, 1837–1850. [CrossRef]
159. Zechman, E.M.; Ranjithan, S.R. Evolutionary computation-based methods for characterizing contaminant
sources in a water distribution system. J. Water Resour. Plan. Manag. 2009, 135, 334–343. [CrossRef]
160. Kim, M.; Choi, C.Y.; Gerba, C.P. Source tracking of microbial intrusion in water systems using artificial
neural networks. Water Res. 2008, 42, 1308–1314. [CrossRef] [PubMed]
161. Liu, L.; Zechman, E.M.; Mahinthakumar, G.; Ranji Ranjithan, S. Identifying contaminant sources for water
distribution systems using a hybrid method. Civ. Eng. Environ. Syst. 2012, 29, 123–136. [CrossRef]
162. Liu, L.; Zechman, E.M.; Mahinthakumar, G.; Ranjithan, S.R. Coupling of logistic regression analysis and
local search methods for characterization of water distribution system contaminant source. Eng. Appl.
Artif. Intell. 2012, 25, 309–316. [CrossRef]
163. Shen, H.; McBean, E. False negative/positive issues in contaminant source identification for water-
distribution systems. J. Water Resour. Plan. Manag. 2011, 138, 230–236. [CrossRef]
164. Eliades, D.; Lambrou, T.; Panayiotou, C.; Polycarpou, M. Contamination event detection in water
distribution systems using a model-based approach. Proc. Eng. 2014, 89, 1089–1096. [CrossRef]
165. Hamam, Y.; Hindi, K. Optimised on-line leakage minimisation in water piping networks using neural nets.
Proc. IFIP Work. Conf. 1992, 28, 57–64.
166. Koch, M.W.; McKenna, S.A. Distributed sensor fusion in water quality event detection. J. Water Resour.
Plan. Manag. 2010, 137, 10–19. [CrossRef]
167. Eliades, D.G.; Polycarpou, M.M. Water contamination impact evaluation and source-area isolation using
decision trees. J. Water Resour. Plan. Manag. 2011, 138, 562–570. [CrossRef]
168. Brumbelow, K.; Torres, J.; Guikema, S.; Bristow, E.; Kanta, L. Virtual cities for water distribution and
infrastructure system research. In Proceedings of the World Environmental and Water Resources
Congress 2007: Restoring Our Natural Habitat, Tampa, FL, USA, 15–19 May 2007; pp. 1–7.
169. Costa, D.; Melo, L.; Martins, F. Localization of contamination sources in drinking water distribution
Systems: A method based on successive positive readings of sensors. Water Resour. Manag. 2013, 27,
4623–4635. [CrossRef]
170. Laird, C.; Biegler, L.; van Bloemen Waanders, B.; Bartlett, R. Time dependent contamination source
determination for municipal water networks using large scale optimization. J. Water Resour. Plan. Manag. 2003.
Available online: https://www.researchgate.net/publication/266407494_T (accessed on 23 November 2017).
171. Laird, C.D.; Biegler, L.T.; van Bloemen Waanders, B.G. Mixed-integer approach for obtaining unique solutions
in source inversion of water networks. J. Water Resour. Plan. Manag. 2006, 132, 242–251. [CrossRef]
172. Laird, C.D.; Biegler, L.T.; van Bloemen Waanders, B.G.; Bartlett, R.A. Time dependent contamination
source determination: A network subdomain approach for very large water networks. In Critical
Transitions in Water and Environmental Resources Management; American Society of Civil Engineers:
Reston, VA, USA, 2004; pp. 1–10.
173. Preis, A.; Ostfeld, A. Contamination source identification in water systems: A hybrid model trees-linear
programming scheme. J. Water Resour. Plan. Manag. 2006, 132, 263–273. [CrossRef]
174. Nazarovs, S.; Dejus, S.; Juhna, T. Modelling water quality in drinking water distribution networks from
real-time direction data. Drink. Water Eng. Sci. 2012, 5, 39–45. [CrossRef]
175. Quesson, B.; Sheldon-Robert, M.; Vloerbergh, I.; Vreeburg, J.G. Acoustic monitoring of terrorist intrusion
in a drinking water network. In Water Distribution Systems Analysis 2008; American Society of Civil
Engineers: Reston, VA, USA, 2008; pp. 1–12.
176. Tabaa, Y.; Medouri, A.; Tetouan, M. Towards a next generation of scientific computing in the cloud. Int. J.
Comput. Sci. 2012, 9, 177–183.
Water 2018, 10, 579
177. Creaco, E.; Pezzinga, G.; Savic, D. On the choice of the demand and hydraulic modeling approach to
WDN real-time simulation. Water Resour. Res. 2017, 53, 6159–6177. [CrossRef]
178. Creaco, E.; Blokker, M.; Buchberger, S. Models for generating household water demand pulses:
Literature review and comparison. J. Water Resour. Plan. Manag. 2017, 143, 04017013. [CrossRef]
c 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open
access article distributed under the terms and conditions of the Creative Commons
Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).