Genetic Algorithms in Water Supply

Application of Genetic Algorithms in Water Supply System Life-

Cycle Analysis
R. Klasinc1, F. Steinman2, S. Šantl2, L. Gosar2, P. Banovec3
1
Graz University of Technology, Dept. of Hydr. Structures and Water Res. Management, Austria∗
2
University of Ljubljana, Chair of Fluid Mech., Faculty of Civil and Geodetic Engineering, Slovenia
3
Hydraulic Research Institute, Ljubljana, Slovenia

Abstract
The present-day water supply services have started to use decision-support tools in order to meet their customers'
supply requirements on one hand and to achieve economic efficiency on the other hand. These tools help the
network management to find the best strategy, which will comprise the majority of influential parameters and
determine the possible outcomes for each future event and intervention. Due to numerous optimising parameters,
which influence the possible outcomes, an optimisation tool called genetic algorithms – GA was developed.

The paper presents the work performed on a small, but complex water supply system, which included building of
a model, model calibration, and an analysis of development perspectives to ensure long term coverage of
customer needs while taking into account the limited resources available. This enabled an advanced analysis of
the existing performance of the water supply system as well as an analysis of different development options.

The work, based on the EPANET and GA library, focused on the problem of multi-objective functions and
interactions. The analysis results can provide a better understanding of the existing system behaviour, more
efficient management, as well as a clearer view of the development needs.

1 Introduction
Harmonisation of interests is of key importance in the area of water supply and management of water supply
systems (Figure 1). On the one hand, there are customers who want sufficient quantity of quality water along
with suitable standards of supply. On the other, there are managers who wish to (and must) run a financially
sustainable business, municipalities, which according to the law have to provide a suitable public service, etc.
[9]. Therefore, the analyses and development of water supply systems in search for an optimal solution integrate
different approaches using computer modelling, which allows better awareness of technical as well as economic
and legal terms (Figure 2).

Figure 1: The relations between actors, issues and areas Figure 2: Searching for an optimal solution
as part of the scope. considering three issues.
From a hydraulic point of view, a water supply system represents a complex system, which should meet the
required standards of supply practically continuously. That is why a thorough estimation is required prior to the

∗
Coresponding author's e-mail: ROMAN.KLASINC@ TUGRAZ.AT

1
 Genetic Algorithms in Water Supply

implementation of certain works, since a modification of a single parameter (e.g. diameter of pipe) can have an
impact on the situation in the entire system (pressures, flows) or may even aggravate the conditions in some
other parts of the system.

Figure 3 presents a life cycle of a certain system. Prior to the realisation of works, an analysis and optimisation
of determined decision variables have to be made. The starting point is a calibrated hydraulic model of the
system [9]. Upon the completion of the construction process, new conditions must be monitored to find out any
new possible problems and interests. To support and manage life cycle processes, an engineer's working
environment and tools were installed [1] according to the scheme illustrated as follows on Figure 4.

Users Environment

Problem/interests CAD
AutoCAD

Analisys and
optimization Monitoring

Modeling Decision DATA Modeling

making WAREHOUSE
MS Access

Realization

GIS
ArcView

Figure 3: Life cycle of a certain system. Figure 4: An engineer's working environment and tools

A lot of work has already been done in the area of water engineering optimisation [8], but the number of decision
variables can be enormous, therefore, the harmonisation of interests and requests may turn into a long-lasting
process. Recently, such problems are tried to be solved with the application of Genetic Algorithms (GA), the
basic idea of which is to imitate the natural evolution processes.

2 Changing Environment of Life-Cycle is modelled with GA

Genetic algorithms (GA or GAs) are a class of non-deterministic algorithms deriving from nature evolution. GA
were originally conceived by John Holland in the 1970’s, and have since been further subsequently developed by
many others [9]. These algorithms are best suited to solving combinatorial optimisation problems, which cannot
be solved by using more conventional operational research methods. Their modus operandi is based on
mechanics of natural selection, combining survival of the fittest with randomised, and yet structured, information
interchange among the individuals in the population.

Over the last 10 years, the number of studies related to the use of GA optimisation in water supply and
distribution problems has considerably increased. One such use of genetic algorithms involves the integration of
the GA approach and the hydraulics solver. This model has proved to be a very effective tool for planning the
optimal strengthening, rehabilitation, calibration, rezoning and operation of distribution networks [8], [6].

For a better understanding of GA as an optimisation tool, Figure 5 presents a case with a simple water supply
system. Decision variables are diameters of the pipes and characteristics of the tank (diameter and water level
elevation). In this case, variables are encoded in integer-number genomes. Further, random initialisation of
population is established in the predefined range of possible solutions.

After the initialisation, the optimisation starts with an evaluation of each genome/individual of the population.
The decision variables are exported into a hydraulic model to calculate hydraulics behaviour of the system. Each
genome is evaluated with regard to the supply requirements (inappropriate pressures are penalised), costs (higher
costs are penalised, too), and some particular requirements (ecology, other plans and interests, etc.).

2
 Genetic Algorithms in Water Supply

Upon the completion of the evaluation of the entire population, the genomes are ranged with regard to the total
score (sum of all scores – requirements, costs, etc.). The worst genomes are eliminated, while the others are put
into the process for the creation of a new (child) generation:

- crossover. The function of crossover is to ensure that the child chromosomes differ from their parents.
Once two parent genomes have been selected (better evaluated chromosomes have more possibilities for
multiplication), crossover involves the exchange of parts of their genetic material to form two new child
genomes;
- mutation. Mutation is final operator in the genetic process and it operates on single chromosomes where
each gene is altered with small probability. If the probability of mutation is too high, the GA degenerates
into random search process [8].

Hydraulic model of planned Construction of chromosome

water supply system with numbering 1 2 3 4 5 6 7 8

Definition and encoding of the optimisation problem

of decision variables
Genes for pipe
Water tank 7 Possible values for tank elevation: decision variables
510, 511, 512, 513, 514, 515 m.s.l.
8 Possible values for tank diameter: Genes for tank
6, 7, 8, 9, 10, 12 m decision variables
Water source
1 2
5
3 6 Initial population of size
4 n - genomes
Possible values for pipe diameter: 1 2 3 ... n-1 n
100, 120, 150, 200, 250, 300 mm. 100 120 250 300 250
100 100 120 200 120
300 150 100 120 200
200 150 150 300 250
Hidraulic Entering values of 100 120 250 ... 100 150
calculation genome into hydraulic 120 150 200 150 120
model 510 512 511 510 514
7 9 12 7 12

Cycle
for each
genome
Optimisation

Valuation of genome:
- supply requirements Generation
Generacija
- costs
- special requirements

Reestablishing of a new child population

Selection for Crossover: - exchange of parts Mutation:
of genetic material
reproduction
120 8
- alternation of genes
Fitter genomes have with some probability
more chances of
reproduction 150 12

Figure 5: Overview of an optimisation process applying GA on simplified Water Supply Network.

The whole process of optimisation is repeated until there are no more changes in the best score. Figure 6 presents
how the best score slowly converges through generations to the optimum solution considering the stated terms
and fitness function.

3
 Genetic Algorithms in Water Supply

25
GA optimisation

The best individual - genome

20
Theoretical optimal solution

15

Score
10

0
0 1000 2000 3000 4000 5000 6000 7000 8000
Generation

Figure 6: Process of GA optimisation is followed with presentation of the score of the best genome in each generation.

3 Case study of WSS Logatec

In order to demonstrate the genetic algorithm

optimisation approach, the water supply system of the
town of Logatec was examined (Figure 7). The system
supplies around 7000 residents and a few industrial
consumptions. It can be placed among old systems,
since some parts are older than 80 years.

Due to different water sources (springs, wells) and

stage development, the system is divided into three
zones: high-pressure, middle-pressure and low-
pressure part. Through regulation elements (pressure
reducing valves, pressure sustaining valves, etc.) the
water can also be distributed from higher to lower
zones.

Figure 7: Skeletonized Water Supply System of Logatec city.

Today, the water supply system of Logatec operates to its limits. During the deficiency of water in the time of
peak consumption, the requirements of supply are continuously disturbed (low pressures). To ensure undisturbed
water supply, the interventions have been planned in the lower-pressure part (Figure 8). The plan includes two
well-pump stations (wells CG-4 and CG-5) with a storage facility (tank Sekirica) and suitable pipeline
connections. The planning must take into account the pressure requirements and water demands as well efficient
management.

Decision variables for improvement 11 Naklo LEGEND:

CG-4 13 of water supply system of Logatec 12 10 9
7 8
CG-4, CG-5…new wells
1
1,2,… n … changes of WSS
6 characteristics
CG-5 14 c. Transportni vo
d
Tabor ka
ž aš 3
Tr
2
16
5

15
4
Sekirica

Figure 8: Detailed area where improvements of WSS were proposed

4
 Genetic Algorithms in Water Supply

In accordance with the proposed interventions from different projects, further decision variables and their
possible values (with widening of the range) have been defined:
- changes of pipe diameters: (genes from 1 to 11) [mm] : 100, 150, 200, 250, 300, 350, 400, 450, or a pipe
is not necessary,
- change of reduction valve diameter (gene 12) [mm] : 100, 150, 200, 250, 300, 350, 400, 450, or a valve is
not necessary,
- changes of pump power: estimated pump curve is multiplied by factors (genes 13 and 14): 0.7, 0.8, 0.9,
1.0, 1.1, 1.2, 1.3,
- tank Sekirica elevation (gene 15) [m] : from 520 to 550 (step 1m)
- additional local (Tabor) increase of energy (pressure) by means of a pump (gene 16) [bar] : from 0.1 to 3
bar (step 0.1 bar), with a possibility of no pump added.

How to incorporate proposed improvements, given with above documented variables, into particular
chromosome (genome), consisted of genes, needed for GA calculations, is shown on the Figure 9.

Figure 9: Decision variables of an optimisation analysis and genome

As starting-point values (pump power, elevation of tank), the proposals from technical documentation (previous
projects and studies) were chosen. With pipe diameters, only the standardised pipe diameters were taken into
account. The pump curves (flow/head) for pumps CG-4 in CG-5 were also taken from previous projects.
Multipliers lower than 1.0 reduce the curve and multipliers higher than 1.0 increase the curve [10].

A multiobjective function needs special settings since it combines many objective functions (requirements,
economy, special terms, etc.) We used weighting of a particular objective function with regard to its significance
(Figure 10). The supply requirements are thus defined as the primary criterion since they have to be fulfilled
although that is not financially sustainable. In case of a cutout of a part of the network, fire, etc., the
requirements do not have to be completely fulfilled, so they can be weighted with lower factors.

The multiobjective function consists of:

- pressure requirements for normal supply measured in 6 representative controlling nodes (Figure 11 – black
circles) under two conditions of the system operation (at peak consumption of water). In the first case,
pumps were shut down, so the whole system was supplied from the Sekirica tank, whereas in the second
case the pumps were open;
- restrictions on the pumped amount of water from wells (CG-4 – 20 l/s, CG-5 – 30 l/s [9]). The genomes,
whose decision variable values showed higher amount of pumped water, got extra penalty;
- construction costs of the pipe network, pump stations and the buster;
- pressure requirements in case of a force majeure. Firstly, a fire situation in the area of Naklo with extra
consumption of water (10 l/s) with duration of one hour (supply only from the Sekirica tank). Secondly, a
cutout of the network in the area of the Tržaška road and in the area of a transportation pipeline (Figure 8),
- requirements for flow velocities,
- oscillation of pressure during a day.

The optimum value (Figure 10) varies due to decision variable (picture generalization). Therefore, in case of
normal supply requirements, the optimum value is at 3.5 bars, and in case of construction costs, it is at zero
value. With weighting a particular objective function with regard to its significance, and total score of an
individual, several scenarios, different boundary (technical, economic, legal) conditions etc. could be calculated.

5
 Genetic Algorithms in Water Supply

Normal supply

Score

Total score of an individual

Pumping restriction

Construction costs

Flow velocities

Pressure of oscillation

Decision variables
Optimum value

Figure 10: Weighting of a particular objective function.

Weighting can be changed also during optimization to get different solutions throughout several evaluations if
different influence of individual objective functions is going to be considered. For given technical, economic and
legal boundary conditions the possible optimal solution achieved by GA are presented on Figure 11.

Measured pressures[bar]:
Z ... supply from reservoir
Best genome score: O ... pump supply
I ... faliure
P ... fire fighting

In this solution this part

of network is supplied
from high-pressure part.

Tabor area is supplied

by local pump for pressure
increase, 2.0 bar.

Figure 11: For different scenarios final result of GA optimisation above parameters of 10 pipes is given.

4 Final remarks
The paper presents the applicability of genetic algorithms, which have proved to be effective in many cases
around the world (planning of renovations of systems, leakage detections, calibrations, etc.). A computer
program named GA-net has been developed to carry out the calibration and optimisation of water supply
systems. GA-net is based on the EPANET program for hydraulic simulations and on the GAlib tool [4] for
performing the GA evaluations.

The range of a calibration problem can be very comprehensive, therefore a simplification skeletonization of the
hydraulic model must be carried out. Besides this, the optimal set of monitoring points must be identified to
reduce their number. The GA provide results which are within reasonable limits of deviation from the real
situation.

6
 Genetic Algorithms in Water Supply

The optimisation with GA can be an exacting piece of work, which requires expert knowledge of the conditions
in the water distribution system and takes into account all the factors which influence the optimal solution
(multiobjective function). GA-net can also help us with an explicit identification of the goals, which are not
always clearly determined. Thus, it directly supports the decision making in the area of water distribution. One
of the keys to the achieved results was a good cooperation and support of the managerial staff of the Logatec
water supply company.

The connection of different tools (like GA-net) has proved to be very useful. There are many new research areas,
which can also be integrated (cost-benefits, risk analysis, definition of importance of areas of system, etc.).

5 References
[1] BANOVEC, P. et al, On integration of water management systems with the system of harmonised
nomenclatures. Proceedings, SEMISOFT, ICID, Bratislava, 1999.
[2] Kleiner, Y. et al, Water distribution network renewal planning, Journ. of comp. in civil engineering, 2000.
[3] EUROPEAN STANDARD, Final Draft prEN 805, 1996.
[4] GAlib, A C++ Library of Genetic Algorithms Components, 1996.
[5] Kompare, B. et al, Using machine learning techniques in the construction of models. Ecol. model. [Print
ed.], 1994, vol. 75/76, str. 617-628.
[6] Savic, D. and Walters, G., Genetic algorithm techniques for calibrating network models, Research report
95/12, University of Exeter, UK, 1995.
[7] Savic D. A. et al., Water Supply and Distribution Systems Modelling and Optimisation (WASYMO),
Research Training Network, 2001.
[8] Schaetzen, W., Optimal Calibration and Sampling Design for Hydraulics Network Models, The thesis for
the degree of Doctor of Philosophy in Engineering, University of Exeter, 2000.
[9] Steinman, F. et al, Genetic Algorithms Supported Planning of Water Supply Systems, Journal of
mechanical engineering, 2001.
[10] Rossman, L. A., EPANET 2 – USERS MANUAL, United States Environmental Protection Agency (EPA),
2000, (http://www.epa.gov/ordntrnt/ORD/NRMRL/wswrd/epanet.html).