HydroLINK - 2021 - Nature-Based Optimization Techniques
HydroLINK - 2021 - Nature-Based Optimization Techniques
HydroLINK - 2021 - Nature-Based Optimization Techniques
Even though formal optimization techniques have been applied to several types of problems in the
water industry, the vast majority of examples from the literature is based on (meta)heuristics, and in
particular those inspired on nature. Their application has been shown quite successful and has allowed
significant boosts in performance while reducing costs. This article provides an overview of nature-
inspired optimization techniques, and briefly discusses a number of case studies in which they have
been applied.
There is already a long history of numerical optimization in the water industry1. Many advances have
been made regarding the solution methods from linear and non-linear programming2 to nature-based
optimization techniques, like genetic algorithms, simulated annealing, particle swarm3, ant colony and
many others. Besides the optimization methods, the design problems have evolved from single to
multiobjective, and from deterministic to stochastic2 and robust approaches. For a thorough overview
of the optimal design of water distribution networks and the applied evolutionary algorithms and
1
Jared L. Cohon David H. Marks (1975) A review and evaluation of multiobjective programing techniques.
Water Resources Research, Volume 11, Issue 2.
2
Lansey, K. E., Duan, N., Mays, L. W., and Tung, Y. K. (1989). Water distribution system design under demand
uncertainty. JWRPM, 115(5).
3
Suribabu, C. R., and Neelakantan, T. R. (2006). Design of water distribution networks using particle swarm
optimization. Urban Water Journal, 3(2), 111-120. doi:10.1080/15730620600855928
metaheuristics, the reader is referred to 4, 5. The application of the methods has been shown quite
successful and has allowed significant boosts in performance of many aspects of systems while
reducing costs. In this article, we present an overview of natureinspired optimization techniques, and
briefly discuss a number of case studies where they have been applied.
Evolutionary based approaches:4 where the underlying iterative process includes perturbation
by crossover and/or random mutation and variously selection and replacement of individuals.
Methods here include evolutionary strategies, genetic algorithms, multi and many objective
evolutionary algorithms, genetic programming and differential evolution.
Swarm intelligence based approaches:3 where the underlying iterative process is based on the
movement and interaction of individual agents working as part of a collective (e.g. a herd, a
flock or a swarm). Methods here include particle swarm optimization, ant colony optimization,
artificial bee colony and many other approaches based on a diverse array of organisms
including fireflies, wolves, herons, fish schools and whales.
Other approaches:7 other sets of metaheuristic optimization methods exist that are perhaps
more distantly related to their natural inspiration but are nonetheless nature-inspired.
Methods such as simulated annealing (inspired by the cooling of metal), chemical reaction
optimization and gravity search are examples of these. Other methods that have a related
mode of operation but are not naturally inspired in the conventional sense include methods
such as tabu search and many variants of local search.
Although each of the above approaches is inspired by a different natural system, it is an open question
as to whether they all represent distinctly separate methods in the exploration of the search space.
However, each method has to balance the tradeoff between exploration and exploitation of the space
of possible solutions, which for most problems in the water industry is very large indeed. Methods that
solely exploit (e.g. hillclimbing) will find mediocre solutions quickly, whereas those that solely explore
(e.g. random search) will eventually find good solutions, but over huge timescales. Nature-inspired
algorithms typically embed exploration through the use of a random or probabilistic operation (e.g.
mutation in evolutionary algorithms, probabilistic path selection in ant colony optimization) and
4
Maier, H. R., Kapelan, Z., Kasprzyk, J., Kollat, J., Matott, L. S., Cunha, M. C., Reed, P. M. (2014). Evolutionary
algorithms and other metaheuristics in water resources: Current status, research challenges and future
directions. Environmental Modelling & Software.
5
Mala-Jetmarova, H., Sultanova, N., and Savic, D. (2018). Lost in Optimisation of Water Distribution Systems? A
Literature Review of System Design. Water, 10, 307. doi:10.3390/w10030307
6
Rechenberg I., Küppers U., Scheel A., Mattheck C., Harzheim L. (1998) Evolution und Optimierung. Bionik.
Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-662-06114-5_14
7
Kirkpatrick, S., C.D. Gelatt Jr. and M.P. Vecchi (1983) Optimization by Simulated Annealing. Science, 13 May
1983, 671-680.
exploitation by conferring some preference on solutions that perform well at the optimization task. A
further feature that characterizes these nature-inspired methods is that they are general-purpose
optimization algorithms that can be applied to many different problems. For most algorithms
application to new problems requires the specification of three elements:
• Representation/encoding: this is the mapping between the features of the problem being
solved and the numeric decision variables that will be optimized by the algorithm. Although
some representations will be straightforward, there are often choices to be made that will
determine the efficacy of the search.
• Fitness or objective function(s): provides an assessment of solution quality in terms of one or
more objectives of the problem being solved. Objectives in water industry network problems
usually include calculations of CAPEX, OPEX, hydraulic constraints such as pressures, velocities
and tank penalties, the coverage or detection likelihood for sensor networks, etc.
• Parameter settings: most algorithms have parameter settings that can affect the efficacy of
the algorithm and can vary on different problems. Common parameter settings include
population sizes, number of iterations, perturbation operator selections and application rates
(evolutionary approaches) and various momentum, velocity and pheromone evaporation
terms (swarm intelligence). These are usually set by rule of thumb or through prior
experimentation, although adaptive methods that set these parameters automatically through
the search are becoming increasingly popular.
The choice of algorithm is often dictated by the number and complexity of the decision variables and
constraints, the computational complexity of the objective function and the number of objectives to
be optimized. Often the computational complexity can be the largest factor and if the problem is
particularly time consuming to solve, on the optimization run may require access to high performance
or cloud computing resources, although many problems can be optimized with modern desktop
equipment.
Nature-inspired optimization algorithms are among the best-known approaches for discovering good
solutions to highly complex large-scale problems in reasonable time and have the potential to
transform the design and operation of the complex assets and systems that characterize the water
industry. This is particularly the case when these methods are combined with other popular AI
methods such as machine learning, where the predictive power of these methods can be coupled with
optimization to yield asset and operations upgrade programmes designed for future system demands.
AI is revolutionizing many sectors and as such, it offers great potential for the water industry as well.
A wide range of problems exists in the water industry for which nature-based optimization algorithms
can provide valuable solutions. We give a generic overview here and discuss a number of case studies
for different fields of application and/or geographies in the following paragraphs. For water resource
management, areas of application include model calibration, choosing sampling locations for
monitoring, and risk-based water supply portfolio planning, the optimization of reservoir operation,
and regulation of the abstraction from different sources for scarcity management. Water pipe
networks, both drinking water distribution and wastewater collection networks have also been
subjected to numerical optimization in numerous cases. Not only their layout and sizing, but also their
subdivision into functional sections and the optimal placement of different types of sensors, including
water quality sensors for contami-nation detection and pressure sensors for leak detection have been
explored.
Climate change, population mobility and urban development in cities necessitates the planning of
major distribution network upgrades and requires a phased approach where changes to population
and demand increase over time. Work in 20078 brought together researchers, consultants and city
planners to develop the water distribution system master plan for the City of Ottawa. This phased
expansion reflected expected population and demand increases over a 25-year planning horizon,
optimized by using an evolutionary algorithm. When the work was carried out, the population of
Ottawa was about 800,000 which has risen to almost one million at the time of writing, highlighting
the extent of the planned urban growth and underlining the accuracy of the projections which
anticipated 1.07 million by 2021. The city’s demand is fed by two water purification plants, Britannia
and Lemieux, located along the Ottawa River, with a combined capacity of 640 Millions of Liters per
Day (MLD) and the East and West Urban Communities are fed by a 1,200 mm transmission feed. A
1,220 mm main also feeds the west portion of the South Urban Community and a 762 mm main feeds
a small area on the east side of the river. There are two major storage reservoirs, one located in the
center of the city and the other in the East Urban Community with storage capacity of 108 ML and 82
ML respectively. There are also two smaller reservoirs in the West and South with storage capacity of
34 ML and 18 ML respectively, and additional elevated storage in the communities.
As with many evolutionary optimization applications, the majority of the work was involved in the
development of representation and objective (fitness) function formulations to enable the
evolutionary algorithm to effectively solve the problem. The representation establishing the link
between the algorithm and problem, provided the options to introduce new infrastructure and
upgrade existing assets. A key element here was the introduction of single variables that combined
logical sets of infrastructure upgrades. An example of this is in the introduction of a new tank; this
must be accompanied by the pipework necessary to connect the tank to the network and so was
established as a single ‘decision’ for the algorithm to take. The introduction of these variables had the
dual effect of increasing the engineering feasibility of the developed solutions and reducing the search
space for the evolutionary algorithm. A single objective function minimized costs (CAPEX and OPEX)
and hydraulic penalties under demand scenarios in 2011, 2021 and the final planning horizon in 2031.
The single objective function required a coefficient to balance the cost and individual hydraulic
components, which allowed the optimization to be tailored towards end-user requirements, although
it would also suit a multiobjective approach. Extensive optimization runs were conducted, and a final
2031 solution was developed at an estimated CAD 402M, including CAD 205M for plant expansions,
CAD 110M for new water mains, CAD 45M for pumping stations and CAD 24M and CAD 17M for
reservoir expansions and elevated tanks respectively. However, the optimization was able to show
only CAD 79M was required to satisfy 2011 demands and a further CAD 152M was required between
2011 and 2021, demonstrating the benefit of using multiple planning horizons within the project.
8
Rogers et al. (2007) Application of optimization technology to the development of a water distribution system
master plan for the City of Ottawa, AWWA Annual Conference and Exposition.
Optimum rehabilitation schemes
The water sector is under growing pressure to deliver service that satisfies customer expectations and
regulatory requirements. Urbanization and growing water demand are putting great stress on ageing
or inadequate infrastructure in many countries. This example9 demonstrates how existing scientific
and engineering knowledge benefited from advances in soft computing analytics to address
deficiencies in a water distribution networks. This network is part of a water distribution network of a
city in the UK. The network has grown over years from a small network to a system that serves 400,000
customers.
The network has two reservoirs, 5 connections that import water from adjacent systems, 1,891 nodes
and 2,462 pipes. The pipes have small to moderate diameter sizes with no major transmission mains
due to the way the system evolved in the past. The existing network is unable to satisfy the recent
growth and projected future demands with adequate pressure. The problem was set as a multi-
objective optimization problem in order to generate a set of optimum rehabilitation schemes that
trade-off between capital investment and system performance.
A two-stage methodology was proposed. In the first stage, using network connectivity and topology,
the system was divided into a number of clusters with stronger internal than external connectivity. In
the second stage, three different problem setting strategies, for optimal rehabilitation, were
considered including: 1 | rehabilitation of pipes within clusters 2 | rehabilitation of feed pipelines,
pipes that connect the clusters with deficiency to other clusters or to the sources, and 3 | rehabilitation
of pipes within clusters and feed pipelines.
Using an undirected graph algorithm of the Gephi tool10, 16 clusters with different degrees of pressure
deficiency were identified for this network. The pipes in the clusters that have no performance issues
and do not participate in water transmission to other areas of the network will have no contribution
towards reducing deficiency in the system. Therefore, they were not considered as candidate pipes for
rehabilitation of the system. A total of 248, 149 and 349 pipes were considered for rehabilitation (as
decision variables) for strategies 1, 2, and 3 respectively. The non-dominated sorting genetic algorithm
II (NSGA-II) was used to generate optimum Pareto-front between total cost and number of nodes with
pressure deficiency for different strategies.
The generated results (Figure 1) based on strategy 3 dominated the results generated by both
strategies 1 and 2. The results of strategy 2 were also compared with those generated based on
considering i) all the pipes as design variables and ii) a subset of pipes (567 pipes) based on the
engineering judgment (water company). The optimum Pareto front generated by strategy 3, again
dominated the results generated based on these two problem settings. The optimum solution with no
pressure deficiency generated by strategy 3, has a total cost of GBP 3.05 million. A solution, with total
cost of GBP 4.15 million with 195 nodes with pressure deficiency, was generated independently by the
9
K. Muhammed, R. Farmani, K. Behzadian, K. Diao, D. Butler (2017) Optimal rehabilitation of water distribution
systems using a cluster-based technique, JWRPM, 143 (7)
10
Gephi (2014). Gephi, an open source graph visualization and manipulation software. https://gephi.org (Jun.
22, 2014)
water company by trial and error. A solution, with a similar number of deficient nodes, on Pareto-front
of strategy 3 has a cost of GBP 1.5 million which is 65% cheaper than the solution generated manually.
Figure 1: Layout of proposed pipes for rehabilitation in different clusters for solution with no deficiency.
Different challenges arise when applying optimization techniques to larger real-world networks: the
computational effort involved in applying numerical optimization techniques to such large networks
and being able to translate practical challenges and constraints to formal problem formulations with
clear objectives, constraints and decision variables. To tackle the first challenge, one might think of
high-performance computing and problem specific variators. Regarding the second one, we have
learned that these types of problems are best solved in an iterative process between researchers and
practitioners, wherein each result is assessed, and the optimization problem is adjusted accordingly to
the gained insights. This approach leads to results that are a perfect fit for what water utilities are
looking for and has the added bonus of providing them with new insights into their own water supply
systems.
An optimization tool has been applied to the rehabilitation of real-life networks in the Netherlands. It
uses (modified) genetic algorithms and NSGAII as optimization methods. Network rehabilitation is
approached as a two-phased problem: (1) the optimal design of the network (so called blueprint or
master plan) and (2) the optimal transition between the currently existing network and the blueprint,
i.e., the rehabilitation timeline.
The design of the network blueprint considers the minimization of costs (a function of the diameter
and length of the new pipes), constrained by minimum pressure requirements and commercially
available pipe diameters and materials. For the rehabilitation timeline both hydraulic (improvement
of current pressure deficiencies) and risk based (reduction of pipe failures, which are a function of pipe
diameter, material and age) objectives have been considered, in combination with a practical aspect
regarding the number of construction sites in each rehabilitation step. A construction site is a cluster
of valve sections where old pipes are replaced by new ones. Water utilities prefer to concentrate
rehabilitation works in a few sites, instead of working in a very disperse manner.
This approach was applied to the water distribution network serving the area of Helmond-Mierlo, with
105,000 inhabitants in the Netherlands11. The network model has about 12,000 pipes. Adding to that
32 commercially available pipe diameters, it is clear how large the solution space for this problem is.
By starting the optimization problem from the current pipe diameter values and using problem specific
variators in the GA, it was possible to effectively explore the solution space.
Figure 2 illustrates the obtained results. The costs for rebuilding entirely the network currently in the
ground would be EUR 41.1M. At the peak demand conditions (maximum demand in the past 10 years)
the 30 m pressure requirement is not met at several nodes of the network. The costs of the optimized
blueprint are significantly lower, at EUR 26.4M. At the same time, the hydraulic performance is
significantly improved: the total pressure deficiency in the network (sum of all pressures below the
required 30 m) is reduced by 97%. Regarding the rehabilitation timeline two Pareto fronts were
obtained: (1) trade-off between maximization of hydraulic performance and number of construction
sites, and (2) tradeoff between the minimization of pipe failures and the number of construction sites.
Figure 2 (b) illustrates one of the identified solutions.
Figure 2: Optimized solution for (A) pipe diameters to minimize costs while guaranteeing adequate network performance given
by different colors, and (B) rehabilitation timeline that maximizes the reduction of pipe failures with a maximum of 10
rehabilitation sites per year (the colors and numbers indicate the year in which the pipes should be rehabilitated, pipes with
the same color are rehabilitated in the same year).
11
Vertommen, I., Laarhoven, K. v., Thienen, P. v., Agudelo-Vera, C., Haaijer, T., & Diemel, R. (2018). Optimal
Design of and Transition towards Water Distribution Network Blueprints. Proceedings.
Network rehabilitation is an opportunity for re-designing, an often organically grown network. The
achieved results prove that numerical optimization techniques can be used in this context. Moreover,
the achieved amount of savings allows the water utility to rehabilitate their networks at a higher rate.
Proactively replacing old and fragile pipes with new ones reduces the risk of pipe failure and thus,
water losses due to leakage.
Moreover, having the initial optimization problem defined and all relevant data organized, it makes it
easy to accommodate different objectives, constraints, and scenarios. In this way, the optimization
problem can be re-run when new information, such as changes in urban development or water
demand, becomes available, making it a very flexible approach.
Both societal events and technological advances have pushed the development of techniques for
online water quality monitoring in drinking water distribution systems since the beginning of this
century. Their purpose is generally to protect customers from incidental and/or intentional drinking
water contamination. The number of online monitoring sensors that can be placed in any system is
always constrained by budgetary limitations. Therefore, methods have been developed to determine
optimal sensor placement12, 13 within a drinking water distribution network. Optimality is, however, a
matter of definitions and requirements. The objectives that have been presented in the literature can
be classified roughly into three categories14, aimed at obtaining information, facilitating utility
response, and mitigating the effects of contamination (Table 1).
Table 1: Rate of increase of potential energy as a function of the lake trophic state with Po = 395,343 W
12
Ostfeld A. et al. 2008 The Battle of the Water Sensor Networks (BWSN): A Design Challenge for Engineers and
Algorithms, Journal of Water Resources Planning and Management 134(6), 556-568.
13
Berry J., Boman E., Riesen L.A., Hart W.E., Phillips C.A., Watson J.P and Murray R. 2010 User s Manual TEVA-
SPOT Toolkit Version 2.4. Office of Research and Development, USEPA 600/R-08/041B.
14
P. van Thienen; B. de Graaf; J. Hoogterp; J. van Summeren; A. Vogelaar (2018) Bounds on water quality
sensor network performance from design choices and practical considerations. Water Practice and Technology
(2018)
Of the three classes of sensor placement optimization objectives, those that are information-oriented
are the simplest to compute12, requiring only a network model (hydraulics and material transport). The
more complex effect-oriented approach has been implemented in the Threat Ensemble Vulnerability
Assessment and Sensor Placement Optimization Tool (TEVASPOT)13; here we present some results of
its application to a network model of part of the network of Vitens, the largest water utility in the
Netherlands. Many simulations were performed to study the relationship between the assumed and
actual response times of the utility (after which water consumption is assumed to cease), on the one
hand, and the performance of the sensor network (reduction in the number of people affected) on the
other. In all cases, the sensor locations were optimized using a genetic algorithm. Some results are
shown in Figure 3. Two important observations can be made: Sensors are useless for event detection
if the utility’s response time is too long and Every additional sensor contributes less to the objective
than its predecessors (the law of diminishing returns).
Figure 3: Performance of optimized water quality sensor networks as a function of number of sensors (n) and utility response
times. Results from 14.
When the water utility elects to consider only practically suitable locations for sensor placement, this
may have a significant effect on the network’s performance as shown in a different study presented in
Figure 4 which presents a comparison between the performance of optimizations (again using a
genetic algorithm) for different sets of uniformly distributed nodes used as potential locations (300,
and all 2,700, respectively). In some cases, the network based on the practical set of potential locations
performs better than one based on a larger number of uniformly distributed nodes. The practical set
may include, by chance, suitable locations that are absent from the uniformly distributed sets. The
best performance is seen when all network nodes are considered as candidate locations (grey curve in
Figure 4). But the main conclusion must be that even though optimizing a sensor network configuration
based on practically available locations results in some performance loss compared to networks in
which sensor placement is not restricted, performing an optimization is still worthwhile.
Figure 4: Water quality sensor network performance for ideal and practically feasible locations. Results from 14.
Conclusions
Several decades of development of ideas, methods and applications have resulted in a myriad of cases
which demonstrate the added value of applying numerical optimization techniques to water industry
problems. Nature-based metaheuristic methods have been and continue to be particularly popular
and successful because of their ability to deal with the scale and complexity that are typical in this field
of application. Nevertheless, the vast majority of system design projects in practice continues to rely
on human designs and expert judgment. This is not to say that the human factor should be taken out
of the equation, rather the opposite: the application of numerical optimization techniques taking into
account the deep domain knowledge of the water industry’s experts holds the potential for
performance increase and cost reduction (both monetary and in terms of environmental impact) in all
these projects. The primary gains for the industry from numerical optimization will come from taking
the step to actually more or less universally applying these methods.
Ongoing development in this area is focused on the development of new and faster formulations of
algorithms, often through the combination of one or more techniques and in the development of
methods that can take advantage of modern CPU and GPU (graphical processing unit architectures).
Other areas of development are learning optimization (hyperheuristic) methods15 and multi-method16
search which combine machine learning and optimization components to create methods that can
adapt to new search space domains on the fly. Real-world applications are being addressed through
the development of many-objective17 and human-in-the-loop18 algorithms that aim to consider the
15
Yates, W. B., & Keedwell, E. C. (2020). Offline Learning with a Selection Hyperheuristic: An Application to
Water Distribution Network Optimisation. Evolutionary computation, 1-24.
16
Raad, D. N., Sinske, A., & Van Vuuren, J. H. (2011). Water distribution systems design optimisation using
metaheuristics and hyperheuristics. ORiON, 27(1).
17
Matrosov, E. S., Huskova, I., Kasprzyk, J. R., Harou, J. J., Lambert, C., and Reed, P. M. (2015). Many-objective
optimization and visual analytics reveal key tradeoffs for London s water supply. Journal of Hydrology.
18
Johns, M. B., Mahmoud, H. A., Walker, D. J., Ross, N. D., Keedwell, E. C., and Savic, D. A. (2019, July).
Augmented evolutionary intelligence: combining human and evolutionary design for water distribution
network optimisation. In Proceedings of the Genetic and Evolutionary Computation Conference.
large number of objectives that characterize real world problems and leverage the domain expertise
of experienced staff. In this way, the people that have always been responsible for the design of
systems and their operation continue to be so, but with a new and very powerful tool in their toolbox.
We are becoming more aware of the uncertainties that exist in the models and data that we are
applying optimization to. In addition to this we observe that the world is changing at an ever quicker
pace, and progressing climate change can be expected to have more changes and con-sequences in
store for us for the rest of the century, both in terms of water availability and demand. Considering
these uncertainties in the present state and future conditions and requirements (discussed in 19), it
becomes urgent to start taking these uncertainties into account in the formulation of our optimization
problems. Academic work on robust and resilient optimization has been presented in the past
decades2, 20 this should become the standard in real world applications as well.
References
1 | Jared L. Cohon David H. Marks (1975) A review and evaluation of multiobjective programing
techniques. Water Resources Research, Volume 11, Issue 2,
https://doi.org/10.1029/WR011i002p00208
2 | Lansey, K. E., Duan, N., Mays, L. W., and Tung, Y. K. (1989). Water distribution system design under
demand uncertainty. Journal of Water Resources Planning and Management, 115(5).
doi:10.1061/(ASCE)0733-9496(1989)115:5(630)
3 | Suribabu, C. R., and Neelakantan, T. R. (2006). Design of water distribution networks using particle
swarm optimization. Urban Water Journal, 3(2), 111-120. doi:10.1080/15730620600855928
4 | Maier, H. R., Kapelan, Z., Kasprzyk, J., Kollat, J., Matott, L. S., Cunha, M. C., Reed, P. M. (2014).
Evolutionary algorithms and other metaheuristics in water resources: Current status, research
challenges and future directions. Environmental Modelling & Software, 62, 271-299. doi:
https://doi.org/10.1016/j.envsoft.2014.09.013
5 | Mala-Jetmarova, H., Sultanova, N., and Savic, D. (2018). Lost in Optimisation of Water Distribution
Systems? A Literature Review of System Design. Water, 10, 307. doi:10.3390/w10030307
6 | Rechenberg I., Küppers U., Scheel A., Mattheck C., Harzheim L. (1998) Evolution und Optimierung.
Bionik. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-662-06114-5_14
7 | Kirkpatrick, S., C.D. Gelatt Jr. and M.P. Vecchi (1983) Optimization by Simulated Annealing. Science,
13 May 1983, 671-680.
19
Peter van Thienen and Dragan Savic (2020) The importance of knowing what we do not know - Uncertainty in
planning, designing and modelling of urban water infrastructure. IWA Digital Water whitepaper, 2020
20
Cunha, M., & Sousa, J. (2010). Robust design of water distribution networks for a proactive risk management.
Journal of Water Resources Planning and Management, 136(2).
8 | Rogers et al. (2007) Application of optimization technology to the development of a water
distribution system master plan for the City of Ottawa, American Water Works Association - AWWA
Annual Conference and Exposition, ACE 2007
10 | Gephi (2014). Gephi, an open source graph visualization and manipulation software.
https://gephi.org (Jun. 22, 2014)
11 | Vertommen, I., Laarhoven, K. v., Thienen, P. v., Agudelo-Vera, C., Haaijer, T., & Diemel, R. (2018).
Optimal Design of and Transition towards Water Distribution Network Blueprints. Proceedings, 2(11),
584.
12 | Ostfeld A. et al. 2008 The Battle of the Water Sensor Networks (BWSN): A Design Challenge for
Engineers and Algorithms, Journal of Water Resources Planning and Management 134(6), 556-568.
13 | Berry J., Boman E., Riesen L.A., Hart W.E., Phillips C.A., Watson J.P and Murray R. 2010 User s
Manual TEVA-SPOT Toolkit Version 2.4. Office of Research and Development, USEPA 600/R-08/041B.
14 | P. van Thienen; B. de Graaf; J. Hoogterp; J. van Summeren; A. Vogelaar (2018) Bounds on water
quality sensor network performance from design choices and practical considerations. Water Practice
and Technology (2018) 13 (2): 328 334. https://doi.org/10.2166/wpt.2018.044
15 | Yates, W. B., & Keedwell, E. C. (2020). Offline Learning with a Selection Hyperheuristic: An
Application to Water Distribution Network Optimisation. Evolutionary computation, 1-24.
16 | Raad, D. N., Sinske, A., & Van Vuuren, J. H. (2011). Water distribution systems design optimisation
using metaheuristics and hyperheuristics. ORiON, 27(1).
17 | Matrosov, E. S., Huskova, I., Kasprzyk, J. R., Harou, J. J., Lambert, C., and Reed, P. M. (2015). Many-
objective optimization and visual analytics reveal key tradeoffs for London s water supply. Journal of
Hydrology, 531, 1040-1053.
18 | Johns, M. B., Mahmoud, H. A., Walker, D. J., Ross, N. D., Keedwell, E. C., and Savic, D. A. (2019,
July). Augmented evolutionary intelligence: combining human and evolutionary design for water
distribution network optimisation. In Proceedings of the Genetic and Evolutionary Computation
Conference (pp. 1214-1222).
19 | Peter van Thienen and Dragan Savic (2020) The importance of knowing what we do not know -
Uncertainty in planning, designing and modelling of urban water infrastructure. IWA Digital Water
whitepaper, 2020
20 | Cunha, M., & Sousa, J. (2010). Robust design of water distribution networks for a proactive risk
management. Journal of Water Resources Planning and Management, 136(2).
doi:10.1061/(ASCE)WR.1943-5452.0000029