Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску
This study introduces a method for determining the best locations for pressure sensors in water supply networks and for assessing network conditions using artificial intelligence techniques. The goal is to identify the network nodes that would provide the most important information for detecting wat...
Saved in:
| Date: | 2026 |
|---|---|
| Main Authors: | , |
| Format: | Article |
| Language: | English |
| Published: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2026
|
| Subjects: | |
| Online Access: | https://journal.iasa.kpi.ua/article/view/365260 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | System research and information technologies |
| Download file: |  |
Institution
System research and information technologies| _version_ | 1869472196250828800 |
|---|---|
| author | Zaychenko, Yuriy Starovoit, Tetiana |
| author_facet | Zaychenko, Yuriy Starovoit, Tetiana |
| author_institution_txt_mv | [
{
"author": "Yuriy Zaychenko",
"institution": "National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv"
},
{
"author": "Tetiana Starovoit",
"institution": "National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, Kyiv"
}
] |
| author_sort | Zaychenko, Yuriy |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2026-06-30T06:14:59Z |
| description | This study introduces a method for determining the best locations for pressure sensors in water supply networks and for assessing network conditions using artificial intelligence techniques. The goal is to identify the network nodes that would provide the most important information for detecting water leaks and evaluating the overall network status. The selection of sensor locations was based on data sets of pressure changes caused by various leak scenarios generated by EPANET simulations. Genetic algorithms were used to rank candidate nodes and determine the optimal number of sensor locations. The next step involved assessing the network state using the ANFIS neuro-fuzzy network and the Mamdani neuro-fuzzy logical inference algorithm. These algorithms were implemented in the Google Colab environment and tested on a section of the water supply network in Kyiv, Ukraine. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2026.2.06 |
| first_indexed | 2026-07-01T01:00:18Z |
| format | Article |
| fulltext |
Yu. P. Zaychenko, T. V. Starovoit, 2026
Системні дослідження та інформаційні технології, 2026, № 2 83
TIÄC
ТЕОРЕТИЧНІ ТА ПРИКЛАДНІ ПРОБЛЕМИ
ІНТЕЛЕКТУАЛЬНИХ СИСТЕМ ПІДТРИМКИ
ПРИЙНЯТТЯ РІШЕНЬ
UDC 004.8:628.1
DOI: 10.20535/SRIT.2308-8893.2026.2.06
HYBRID COMPUTING INTELLIGENT SYSTEM FOR
ASSESSING THE STABILITY OF THE WATER DISTRIBUTION
SYSTEM AND DETERMINING THE OPTIMUM LOCATIONS
OF PRESSURE SENSORS
YU.P. ZAYCHENKO, T.V. STAROVOIT
Abstract. This study introduces a method for determining the best locations for
pressure sensors in water supply networks and for assessing network conditions using
artificial intelligence techniques. The goal is to identify the network nodes that would
provide the most important information for detecting water leaks and evaluating the
overall network status. The selection of sensor locations was based on data sets of
pressure changes caused by various leak scenarios generated by EPANET
simulations. Genetic algorithms were used to rank candidate nodes and determine the
optimal number of sensor locations. The next step involved assessing the network
state using the ANFIS neuro-fuzzy network and the Mamdani neuro-fuzzy logical
inference algorithm. These algorithms were implemented in the Google Colab
environment and tested on a section of the water supply network in Kyiv, Ukraine.
Keywords: sensor placement, WDN, artificial intelligence, hydraulic modeling,
ANFIS, Mamdani, genetic algorithms, EPANET 2.2.
INTRODUCTION
A water supply network (WDN) is a complex engineering network made up of
pipelines and other elements that ensure the functioning of the water supply system
from the source to the end users (end nodes). The primary function of the WDN is
to deliver the expected volume of water at sufficient pressure. Real-time monitoring
of the network’s condition is crucial for stable operation to assess its current
performance and ability to fulfill its core functions.
Most studies focus on quantifying the performance of the entire WDN. Shin
et al. [28] proposed quantitative methods for determining the stability of the water
supply network. Analyzing the network at the global level using a single metric
does not provide information about the level of performance of the nodes in the
network. Therefore, it is important to monitor the performance of each network
node. The pressure information at each node and the corresponding nodal outflow
information are necessary to evaluate the state of the network at the node level.
It also helps in adaptive network management for increased returns and maximum
quality of service. It follows that data collection from each demand node is
necessary. However, it is not possible to install measuring devices throughout the
network. Since the more sensors, the higher the cost of initial installation and
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 84
maintenance of the network. There is also a need to power the sensors for their
operation.
Determining the ideal placement for pressure sensors is a key challenge for
monitoring water distribution networks. It’s not feasible to install sensors at every
point within the network, which consists of thousands of nodes. Therefore, only
a select few nodes can have sensors installed. The main goal is to figure out the best
locations for these sensors so that they can provide relevant data for assessing
hydraulic variables at unmonitored points and help with various monitoring
algorithms, such as leak detection [23]. However, the data from these sensors alone
might not be sufficient to accurately pinpoint leaks and evaluate the network’s
condition, so additional sensors may need to be installed in other areas [24]. One
solution to this problem is to install more pressure sensors, as they are more cost-
effective and easier to install and maintain compared to flow rate sensors.
Furthermore, pressure readings are more sensitive to leaks than flow rates, so many
localization algorithms rely on network pressure measurements.
Once a sufficient number of measurements are obtained, they can be used to
estimate the unmeasured states of the WDN at any point in time. The problem of
restoring missing information or assessing the state of the system from damaged
data is not new. There are several methods based on spatial, spectral, temporal, and
statistical approaches. Hu et al. [29] worked on the estimation of discrete-time
networks. In this work, it is assumed that the measurements follow a certain
distribution. In the study [30], state assessment was performed for networks with
a variable topology. Liu et al. [31] developed a decentralized state estimator for
spatially distributed systems. The implementation involves state space matrices
A and C, which have the shape of a block diagonal. This method of matrix filling
(MS) for state estimation is quite popular in various fields.
In order to enhance network reliability, Chandramouli and Malleswarvrao
[11] utilized fuzzy logic based on excess pressure available at demand nodes.
Prasad and Park employed genetic algorithms, taking into account both cost
minimization and network reliability maximization [12]. Recent advancements
include improving algorithm convergence by employing a constructed starting
population instead of a random one [13], enhancing computing efficiency by
reducing the search space [14], and combining GA and mathematical programming
with the inclusion of new elements, such as reduction valves [15]. Other methods
involve the use of artificial neural networks (ANNs) instead of hydraulic and
simulation models for water quality, along with differential evolution (DE) for
optimization [16], as well as the development of the Harris Hawks optimization
algorithm (HHO) for WDN optimization [16].
While gathering information about the water supply network in Kyiv, we
encountered difficulties due to limited and unstructured data, as well as a large
amount of paper records. To address this, active efforts are underway to digitally
transform and develop GIS systems. In this study, Mamdani’s method [22],
a reliable and flexible fuzzy logic inference method, was adapted to convert non-
quantitative “expert” knowledge into a quantitative scale. This adaptation is
valuable for mapping a relatively complex water supply network and allows for the
integration of non-statistically independent components.
This paper proposes a method for determining the optimal locations of
pressure sensors using genetic algorithms and assessing the state of the water
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 85
supply network using the ANFIS neuro-fuzzy network and the Mamdani fuzzy
logic inference algorithm. The results of this study make it possible to assess the
state of the water supply network and find the best places for the placement
of pressure sensors. Based on the previous research of Khorshidi [16], who also
performed the relevant work, this method not only helps to find the leakage points
but also evaluates the state of the water supply network.
The proposed intelligent model can be used for controlling pressure or flow
rate sensors in water drainage networks. It is necessary to utilize software tools for
hydraulic modeling of water drainage networks, as pipeline breaks are also present
in sewage networks [18–20]. Additionally, this method can help identify the best
locations for water quality monitoring sensors. With the assistance of neuro-fuzzy
neural networks and fuzzy logical inference algorithms, it can also be used to assess
network conditions and predict water quality.
DESCRIPTION OF MODEL DEVELOPMENT AND TRAINING
METHODOLOGY
Modeling of water demand
The cor-PRP model, which was proposed in the study [2], was used to form the
demand at the common node. In this model, the nodal frequency of water use in
residential premises corresponds to a non-stationary Poisson process. In particular,
the probability that Pr has exactly k pulses generated during a time period of
duration ∆τ[c] is determined by the following equation: 𝑃𝑟(𝑘) = 𝑒 ∆ (𝜆∆𝜏)𝑘! , (1)
where the parameter λ[c-1] is the average pulse arrival frequency, which is
considered constant during ∆τ. After generating pulse arrivals, the duration D [s]
and intensity I [l/s] of the total pulses are sampled from probability distributions
such as the log-normal and beta distributions [2]. An important feature of the cor-
PRP model is the correlation between pulse duration and intensity, which improves
pulse stability [2].
Hydraulic model of the water supply network
A hydraulic model is commonly used to calculate hydraulic parameters such as
water pressure and flow rate for water network design. Hydraulic formulas describe
the conservation of mass and conservation of energy, taking into account the
topological characteristics of the water supply network. The hydraulic model takes
into account fluctuations in water demand and leaks that affect network
performance. The main formulas used for hydraulic modeling are indicated in the
equations below [4].
Formula (2) calculates the mass transfer in the pipe node; it indicates that in
the absence of leakage, the inflow of water to the pipe node should be equal to the
outflow of water [4]: 𝑞 ,∈ − 𝐷 = 0, ∀𝑛 ∈ 𝑁 , (2)
where 𝑃 is a set of pipes connected to node n; 𝑞 , is the flow of water in node n
from pipe 𝑝 (𝑚 /𝑠), 𝐷 . Formula (3) describes the energy transfer, that is, the
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 86
total water head, which includes components describing kinetic energy (kinetic
water head), hydraulic potential energy (head) and gravitational potential energy
(height head) [4]: ℎ = 𝑢2 + 𝑝𝛾 + 𝑧 = ℎ + 𝐻 = 𝑢2 + 𝑝𝛾 + 𝑧 + 𝐻 , (3)
where h is the total water head; u is the water velocity at each node; and z is the
height of each node. 𝐻 is the energy loss value between node A and node B [4].
Energy consumption in the pipe flow can be distributed or localized. The
distributed energy consumption is determined by the flow rate V, the internal
diameter of the pipe d, the length of the pipe L, and the roughness of the pipe wall,
which is determined by the Hazen–Williams formula [5], formula (4): 𝐻(𝑚) = 6.78𝐿𝑑 . (𝑉𝐶) . , (4)
where C is the roughness coefficient of the pipe wall. Localized energy losses occur
due to turbulence associated with changes in flow conditions (such as flow velocity,
direction, etc.) determined by the topology of water supply network connections [4].
For the water supply network, the main thing is the consumption or demand
for water. Two models are used for water demand in nodes: a demand-driven model
and a pressure-driven model [4]. This study uses a pressure-driven water demand
model to consider the effects of pressure loss due to changes in water demand or
leaks [4]:
𝐷 = ⎩⎨
⎧ 0𝐷 ( 𝑝 − 𝑃𝑃 − 𝑃 )𝐷 𝑝 ≤ 𝑃 𝑃 ≤ 𝑝 ≤ 𝑃 , 𝑝 > 𝑃 (5)
where D is the demand in each node; 𝐷 is the desired demand (𝑚 /𝑠); p is the
water pressure; 𝑃 is the pressure above which the desired demand 𝐷 must be
satisfied; 𝑃 is the pressure below which water will not be supplied to the node [4].
Genetic algorithms of sensor placement
In the water supply network, the main focus is on water consumption or demand.
There are two models used for water demand in nodes: a demand-driven model and
a pressure-driven model [4]. This study utilizes a pressure-driven water demand
model to account for the impacts of pressure loss caused by variations in water
demand or leaks [4].
Each possible solution of the optimization problem, using GA, is called
a chromosome. The mathematical formulation of the optimization problem is based
on the fact that each chromosome consists of a series of genes (decision variables)
that represent a possible solution to the optimization problem. In an N-dimensional
optimization problem, a chromosome is an array of size 1 × N. This array is defined
as follows [3]: 𝐶ℎ𝑟𝑜𝑚𝑜𝑠𝑜𝑚𝑒 = 𝑋 = (𝑥 , 𝑥 , … , 𝑥 , … , 𝑥 ),
where X is a possible solution to the optimization problem; 𝑥 is the i-th decision
variable (or gene) of decision X; and N is the number of decision variables.
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 87
A genetic algorithm begins by randomly generating a population of chromosomes
or possible solutions. The size of the population, or the number of possible
solutions, is presented in the form of a matrix of chromosomes of size M × N [3]:
𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 =
⎣⎢⎢
⎢⎢⎡
𝑋𝑋⋮𝑋⋮𝑋 ⎦⎥⎥
⎥⎥⎤ =
⎣⎢⎢
⎢⎢⎡
𝑥 , 𝑥 , ⋯ 𝑥 , ⋯ 𝑥 ,𝑥 , 𝑥 , ⋯ 𝑥 , ⋯ 𝑥 ,⋮𝑥 , 𝑥 , ⋯ 𝑥 , ⋯ 𝑥 ,⋮𝑥 , 𝑥 , ⋯ 𝑥 , ⋯ 𝑥 , ⎦⎥⎥
⎥⎥⎤,
where 𝑋 is the j-th solution (or chromosome); 𝑥 , is the i-th solution variable
(or gene) of the j-th solution; and M is the population size. Each decision variable 𝑥 , can be represented as a floating-point number (real values), or as a predefined
set of values for discrete problems. Some of the initially generated possible
solutions are selected as parents to create a new generation [3].
Selection in GA is a procedure by which R (R < M) individuals are selected
from a population for reproduction. The selected individuals are the parents of the
next generation and make up the parental population. The ranking option ranks all
chromosomes based on their match values. The best solution gets rank 1, and the
worst gets the lowest rank. The decision is assigned a probability that is
proportional to its rank according to the following linear function [3]: 𝑃 = 𝑈 − (𝑆 − 1) 𝑍 , (6) 𝑆 = 𝑅𝑎𝑛𝑘 (𝑋 ), (7)
𝑃 = 1, (8)
𝑈 = 𝑍 (𝑀 − 1)2 + 1𝑀 , (9)
where 𝑆 is the rank of the k-th solution in the population; 𝑆 = 1 indicates that the k-
th solution is the best solution; and Z is a user-defined value. Fig. 1 shows the sorting
of solutions according to the fit function (F) in the maximization problem [3].
Fig. 1. Ranking of chromosomes according to the correspondence function (F) [3]
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 88
An alternative method ranks all solutions according to their fitness values.
Then, M−S copies of each solution are generated [3].
Pressure sensor placement method using genetic algorithms
The proposed sensor placement method is based on a nodal pressure data set that
simulates typical variations due to leaks of different sizes at all network nodes. Data
on pressure in the network were obtained using hydraulic modeling of the water
supply network of the city of Kyiv. Each pressure data point is labeled with a leak
class, for further classification of the data.
The method of placement of pressure sensors is the stage of selecting
a function. In order to select the characteristics (a subset of nodes where the sensors
will be placed), an algorithm is proposed that seeks to maximize the relevance of
the selected characteristics (node pressure) for the response variable (leakage
node). Each response variable avoids capturing information already contributed by
others, that is, redundancy is minimized. The definitions of relevance and
redundancy proposed in the study [32] are used as a basis for determining the
methodology:
a) Relevance is an indicator of the relevance of a subset of nodal pressures. It
is calculated according to the formula: 𝑅𝑒𝑙 = (𝒮) ≝ 1𝑆 𝐼 ( ∈ 𝒮 𝑥, 𝑦) , (10)
where x is any function in 𝒮; and 𝑆=|𝒮|is the number of functions in 𝒮 (power).
b) Redundancy – the metric of information redundancy in a subset of the
function 𝒮, determined by the formula: 𝑅𝑒𝑑 = (𝒮) ≝ 1𝑆 𝐼 (, ′ ∈ 𝒮 𝑥, 𝑥 ′) , (11)
where x and 𝑥′ are the quality of the feature in 𝒮.
To apply the above formulas in order to calculate the location of the pressure
sensor, a data set of pressures at the nodes is first created, which provides different
scenarios that take into account leaks of different sizes at all nodes of the network.
A series of nodal pressure samples, one sample for each leakage scenario, were
obtained using hydraulic simulation. If M different leakage scenarios are simulated
in a network containing N nodes, the result of the simulation is a set of
МN-dimensional vectors, x and 𝑥′ in (10) and (11), corresponding to N candidate
nodes (it is assumed that all nodes are potential definition nodes). Additionally, an
output vector y (Equation (10)) containing integers is generated to indicate the leak
node corresponding to each simulated scenario.
Finding the optimal subset of sensors 𝒮 requires testing 2N different
combinations, which would require impractical computation time in multi-node
networks. Therefore, the proposed method [32] is used to rank pressures in nodes
using an iterative forward scheme that requires only (𝑁𝑆) combinations. With the
help of this scheme, you can rank all the pressures in the nodes in order of
importance with the calculated costs (𝑁2).
The next step involves creating a genetic algorithm to rank node pressures
based on their importance in identifying different classes of leaks (nodes with
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 89
leaks). The output list will start with the nodes considered most important for
identifying leak locations based on the information in the dataset. The process of
selecting nodes will begin with an empty subset and, at each step, the node with the
best ranking among those not yet selected will be added. Each iteration will
evaluate the relevance of each available function (nodal pressure) in relation to the
output (leak node) and its redundancy with respect to previously selected variables.
The number of sensors needed for pressure monitoring and leak detection
depends on the available equipment. The goal is to minimize the number of sensors
based on the method’s use of available information, measurement noise, sensor
calibration quality, and resolution. It’s important to consider that increasing the
number of sensors doesn’t always result in better outcomes. The process involves
starting with one sensor (the best-ranked one) and gradually increasing the number
of sensors while evaluating the leak localization performance for each new set of
sensors. This continues until adding a new sensor no longer provides a significant
advantage.
Intelligent models for assessing and forecasting the state of the water network
Mamdani Fuzzy Logic Inference System (MFISM). The model of fuzzy logic
was presented by Zadeh in 1965 [33] in order to create systems close to human
thinking. Works [34, 35] proved that this method is effective in the development of
complex systems with uncertain conditions [37].
Fuzzy Logic Inference (FIS) consists of three components: fuzzification,
inference, and defuzzification. In this system, knowledge is represented as a set of
fuzzy linguistic rules that allow making fuzzy decisions (not 0 or 1, but values from
0 to 1). The human expert can be replaced by a combination of a fuzzy rule-based
system (FRBS) and a block called a defuzzifier [37].
Fuzzification is a process of decomposition of input and output data into one
or more fuzzy sets. The fuzzy set is defined using a membership function that
represents the region of interest, i.e., the main interests in the interval [0, 1]. The
shape of the curves shows the membership function for each data set and can be
expressed in various geometric shapes, such as: trapezoid, triangle, etc. The
membership function represents the degree of belonging of the values in the data
set [36].
The membership function of the set A defined on the domain X has the form:
μ(A): X [0; 1], the set A is defined through its membership function µ by the
equation [37]: 𝜇 (𝐴)𝜖 = 1, 𝑖𝑓 𝑥 𝑖𝑠 𝑓𝑢𝑙𝑙 𝑚𝑒𝑚𝑏𝑒𝑟 𝑜𝑓 𝐴∈ 0, 1 𝑖𝑓 𝑥 𝑖𝑠 𝑝𝑎𝑟𝑡𝑖𝑎𝑙 𝑚𝑒𝑚𝑏𝑒𝑟 𝑜𝑓 𝐴 = 0, 𝑖𝑓 𝑥 𝑖𝑠 𝑛𝑜𝑡 𝑚𝑒𝑚𝑏𝑒𝑟 𝑜𝑓 𝐴 . (12)
The following set for the trapezoidal membership function f is calculated
according to the formula [37]:
𝑓 (𝑥, 𝑎, 𝑏, 𝑐, 𝑑) =
⎩⎪⎪⎨
⎪⎪⎧ 0, 𝑖𝑓 𝑥 < 𝑎 𝑜𝑟 𝑥 > 𝑑(𝑎 − 𝑥)(𝑎 − 𝑏 ) 𝑖𝑓 ≤ 𝑥 ≤ 𝑏1, 𝑖𝑓 𝑏 ≤ 𝑥 ≤ 𝑐(𝑑 − 𝑥)(𝑑 − 𝑐) , 𝑖𝑓 𝑐 ≤ 𝑥 ≤ 𝑑 . (13)
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 90
Inference rules represent relationships between subsets of inputs and outputs.
Inference rules should create a new output subset. Each rule consists of two parts:
“If” and “Then” [37].
The last stage of processing is the defuzzification procedure. This process
allows you to convert the results obtained as fuzzy sets into numerical values. The
center of gravity, the average of the maxima, and the smallest of the maxima are
widely used as defuzzification methods [38].
Adaptive neuro-fuzzy logical inference system (ANFIS). The adaptive
neuro-fuzzy inference system consists of five different levels. The first level is
responsible for identifying input data and output variables and defining their
descriptors. The second level defines membership functions for each input and
output variable. The third level creates a rule base. Level 4 performs rule
evaluation. The last level (Level 5) performs defuzzification [39].
At the first level, each “i” node of this equation is an adaptive node and a node
membership function (MF) [39]: 𝑂 = 𝜇 (𝑥), 𝑖 = 1, 2, 3 … (14) 𝑂 = 𝜇 (𝑥), 𝑖 = 1, 2, 3 … (15)
Fuzzy MFs have different shapes, such as Gaussian, triangular, and
trapezoidal.
Level 2. Calculation of power of rules [39]: 𝑂 = 𝑊 = 𝜇 (𝑥) ∗ 𝜇 (𝑥), 𝑖 = 1, 2, 3 … (16)
Level 3. Normalization of calculated values [39]: 𝑂 = 𝑊∑ 𝑊 , 𝑖 = 1, 2, 3, 4 … (17)
Level 4. In this level, each node represents a sequential part of a fuzzy rule [39]: 𝑂 = 𝑊 ∗ 𝑓 = 𝑊 ∗ (𝑃 ∗ 𝑥 + 𝑞 ∗ 𝑦 + 𝑟 ) 𝑤ℎ𝑒𝑟𝑒 𝑖 = 1, 2. . . 𝑛 . (18)
At level 5, the sequential part of the rules is defuzzified by summing the
outputs of all rules [39]: 𝐹𝑖𝑛𝑎𝑙 𝑂𝑢𝑡𝑝𝑢𝑡 = 𝑂 = 𝑊 ∗ 𝑓 = 𝑊 ∗ (𝑃 ∗ 𝑥 + 𝑞 ∗ 𝑦 + 𝑟 ). (19)
SIMULATION RESULTS
Determination of the number of pressure control points and optimum
placement scheme
For this study, the hydraulic zone of the digital double of the water network of the
city of Kyiv (Fig. 2) with a length of approximately 30 km, 1335 connections and
an average input flow of about 65 m3/h was used. A hydraulic simulation model
was developed in EPANET 2.2 [40], with an hourly demand pattern. The nodal
flow was divided into water consumption by the user and leakage. To establish the
parameters of the model, a relationship between leakage and pressure was
constructed.
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 91
Fig. 2. Hydraulic zone of the network of the city of Kyiv (Ukraine)
To construct the pressure data set, leaks of different magnitudes were
simulated at each connection node using the hydraulic modeling program EPANET
2.2 [40]. The process of creating a dataset using the EPANET program, training
and predictive use of classifiers for leak detection are described in [41]. The data
set generated by simulations for this study considered leaks at all connection nodes
with flow rates starting at 50 L/s. To simulate nodal leaks, the demand assigned to
a given node in the EPANET hydraulic model was modified by increasing the
demand by an amount equal to the flow of the simulated leak. The maximum
number of installed sensors, 𝑁 , was determined based on the characteristics of
the network, namely the length of the network (one sensor per km).
As a result, two pressure sensitivity matrices were calculated. Pipe roughness
indicators were determined by the Hazen–Williams formula (20) (for the first
matrix): 𝐹(𝑁) = (𝑎𝑁)(𝑏 + 𝑁) , (20)
where a, b, c and d are function parameters.
The second matrix was obtained by generating a burst of fixed size for each
node of the hydraulic model with a single emitter factor of 0.25; hydraulic modeling
was performed in EPANET 2.2 [40].
The optimal placement of pressure sensors for a given quantity was
formulated as an unconstrained multi-objective optimization problem. The decision
variables were the nodes where the pressure sensors could potentially be installed, and
all nodes were considered as possible placement locations. The objective functions
were aimed at maximizing the sensitivity of the nodal pressure both to the variations of
the pipe roughness coefficient (𝑓 ) and to the cases of pipe rupture (𝑓 ).
The problem was solved once for each number of sensors in each of the five
discrete sets. The algorithm was implemented in the Google Colab environment on
Python, using the Pymoo package [42]. Each discrete location (i.e., node) was
translated as an integer value. A population member is a set of pressure sensor
locations, and each population member variable represents a possible pressure
sensor location. Fig. 3 shows the general diagram of nodes and edges (joints and
pipes).
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 92
Fig. 3. General scheme of the network
The following parameters were used to develop the GA method: random
integer sampling and selection operators, integer mutation with probability 𝑝 = 0.05, and index parameters 𝑛 = 20. The population size (100) was
considered, and all operations were conducted for 500 generations. This solution
resulted in 500 × 100 = 50,000 objective function evaluations. A PC with an
Intel(R) Core(TM) i5-9400 CPU @ 2.90GHz 2.90 GHz and 16GB of memory was
used for this work with a total running time of ≈ 1 hour. The results of the proposed
options for placement of pressure sensors and the final placement scheme are
shown in Figs. 4–6, respectively.
Fig. 4. Proposed variant 1 of the location of pressure sensors
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 93
Fig. 5. Proposed variant 2 location of pressure sensors
Fig. 6. The final proposed arrangement of pressure sensors
Figs. 4, 5 show that the calculated locations of the pressure monitoring sensors
do not exhibit geometric regularity, as no geometric or spatial criteria were used to
distribute the sensors in the network. However, despite the geometric irregularity,
testing of the leak locations at these locations showed that pressure measurements
at these nodes provided the most useful information for distinguishing between
different leak scenarios. This can be explained informally: “Two teams of players
cannot achieve similar results using different players”. There can be different
sensor placements that provide high performance in leak detection. The next section
describes the evaluation of the network condition using the results described above.
Assessment and forecasting of the state of the water network of the city of Kyiv
Adaptive neuro-fuzzy network (ANFIS). The main advantage of the adaptive
neural network-based fuzzy logic inference system over other systems is that the
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 94
parameters used for its membership function can be changed. The first 8 rules for
each membership function are shown in Figs. 7, 8. Therefore, we can apply
machine learning (ML) algorithms to train these parameters and build a model that
fits the given data set.
Fig. 7. Graphic representation of rules no. 1, no. 2, no. 3, and no. 4 for the membership
function
Fig. 8. Graphic representation of rules no. 5, no. 6, no. 7, and no. 8 for the membership
function
The input parameters of ANFIS are: flow rate, pressure, material, diameter,
material resistance, data on the height of the pipeline location, volume of water
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 95
consumption, year of laying the pipe, type of soil. The data with the selected
parameters were first transformed into the number of principal components and
then loaded as input data into ANFIS using a Gaussian membership function of the
input parameters. The performance results of ANFIS neural fuzzy network training
on the test and training data sets are shown in Table 1.
T a b l e 1 . Assessment of the accuracy of the ANFIS model
No. Test Train
1.
Epoch 200
Loss 0.28776 0.170298
Accuracy 0.89362 0.94715
F1-score 0.87805 0.94222
Precision 0.9 0.91379
Recall 0.85714 0.97248
2.
Epoch 400
Loss 0.31423 0.13415
Accuracy 0.914893 0.951219
F1-score 0.90476 0.946903
Precision 0.90476 0.91453
Recall 0.90476 0.98165
3.
Epoch 600
Loss 0.34790 0.12462
Accuracy 0.91489 0.95121
F1-score 0.90476 0.94737
Precision 0.90476 0.90756
Recall 0.90476 0.99082
4.
Epoch 800
loss 0.37367 0.12151
Accuracy 0.91489 0.95528
F1-score 0.90476 0.95196
Precision 0.90476 0.90833
Recall 0.90476 1.0
5.
Epoch 1000
Loss 0.40200 0.11145
Accuracy 0.91489 0.95528
F1-score 0.90476 0.95154
Precision 0.90476 0.91525
Recall 0.90476 0.99082
For model training, 14 general fuzzy rules were set, and binary cross-entropy
was used as the loss function. During training, we used the Adam optimizer with
the parameter α = 0.01, for 1000 epochs. Data points for plotting were taken from
every tenth epoch. A Gaussian membership function was used for each fuzzy rule.
Weights µ, σ were randomly initialized from a normal distribution. Graphical
results of model performance indicators are shown in Fig. 9.
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 96
Fig. 9. Graphical visualization of ANFIS accuracy assessment
The obtained results indicate that the adaptive neuro-fuzzy model is
successfully trained and works well.
The result of Mamdani’s neuro-fuzzy logic inference method. A key
feature of Mamdani’s algorithm is its ability to adapt to transfer non-quantitative,
“expert” knowledge into a quantitative scale that is useful for mapping a relatively
complex water supply system. This algorithm makes it possible to integrate
components that are not statistically independent into a single index value – this
makes it possible to fully use all available data. An example of the transformation
of data on the location of pipes in the area and data on network failures is shown
in Figs. 10, 11.
Fig. 10. Transformation of “expert” knowledge about the location of the network into
a quantitative scale
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 97
Fig. 11. Example of transmission of data about failures on the network in a quantitative scale
The Mamdani method consists of three steps:
Step1: fuzzification – input values are transformed into categorical variables
(for example, “high”, “low”) using membership functions;
Step 2: fuzzy logical inference – a system of rules is defined that defines an
aggregated function based on the categorical values attributed to an individual
component;
Step 3: Defuzzification – the final value of the index is determined based on
the geometry of the aggregated function. The accuracy results of the Mamdani
algorithm are described in Table 2.
T a b l e 2 . Evaluation result of Mamdani’s neuro-fuzzy logic inference algorithm
No. Metrics Result
1. Accuracy 0.47773
2. Roc-auc score 0.50365
3. F1-score 0.27528
Hence, Mamdani’s neuro-fuzzy logic inference algorithm is well-suited for
translating non-quantitative, “expert” knowledge into a quantitative scale that is
useful for mapping a relatively complex water supply system. Since large amounts
of data on water distribution networks are non-quantitative or approximate.
The assessment of the state of the water supply network of the city of Kyiv
was carried out under normal and abnormal conditions. Performance evaluation at
nodes is important to determine the performance variation between different parts
of the network, which requires the selection of critical regions (nodes) for further
operation and maintenance. The main factors that lead to the deterioration of the
performance of nodes are their location – if the nodes are located at a high altitude,
they have low pressure and flow. If the nodes are far from the treatment station,
they have a lower concentration of residual chlorine (other cleaning reagents).
Thus, nodes with low productivity cannot provide sufficient quantity and quality of
water for the user at a given time, which minimizes the provision of consumer
needs. In addition, there is a need to classify nodes according to the level of
performance.
CONCLUSIONS
In this study, a hybrid computing intelligent system is proposed for assessing the
stability of the water distribution system and determining the optimal locations
of pressure sensors. The results of the study showed that:
the application of artificial intelligence methods to the field of water resources
management indicates a great informational potential, which makes it possible to
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 98
control water supply systems in real time, to automate and implement revolutionary
new methods of analysis and forecasting of the state of engineering networks;
fuzzy forecasting and network condition assessment models have a significant
advantage as they require less information about water supply systems than conventional
probabilistic models. In addition, this information may be vague and inaccurate;
those nodes that received minimum pressure (less than 20 m) and
maximum pressure (more than 50 m) during all simulation periods have low
productivity and negatively affect the quantity and quality of water provided to the
consumer. There are several options recommended for increasing productivity in
critical areas (nodes). For example, replacing pipes, near nodes, laying parallel
pipes, building new tanks for emergency sources of water and chlorine;
in the failure state, the results of the reliability calculation show that the
network can perform its function, that is, the ability to provide a sufficient amount
of water at the desired pressure and water quality by 64.81 %, which has a sufficient
level of performance. Similarly, the resilience analysis shows that the network has
a 53.21 % probability of quickly meeting demand after an event of insufficient
supply, low pressure or a quality that has a high-performance level. In addition, the
research results show that the network has a 17.5 % susceptibility to failure, which
is in the medium vulnerability range.
REFERENCES
1. X. Zhou et al., “Deep learning identifies accurate burst locations in water
distribution networks,” Water Research, vol. 166, 115058, 2019. doi:
https://doi.org/10.1016/j.watres.2019.115058
2. E. Creaco, M. Blokker, S. Buchberger, “Models for generating household water
demand pulses: Literature review and comparison,” Journal of Water Resources
Planning and Management, 143(6), 04017013, 2017. doi:
https://doi.org/10.1061/(ASCE)WR.1943-5452.0000763
3. Omid Bozorg-Haddad, Mohammad Solgi, Hugo A. Loáiciga. Meta-heuristic and
Evolutionary Algorithms for Engineering Optimization. John Wiley & Sons, Inc.,
2017. doi: https://doi.org/10.1002/9781119387053
4. C.P. Liou, “Limitations and proper use of the Hazen–Williams equation,”
Journal of Hydraulic Engineering, 124(9), pp. 951–954, 1998. doi:
https://doi.org/10.1061/(ASCE)0733-9429(1998)124:9(951)
5. Xudong Fan, Xijin Zhang, Xiong ( Bill) Yu. “Machine learning model and strategy
for fast and accurate detection of leaks in water supply network,” Journal of
Infrastructure Preservation and Resilience, vol. 2, article no. 10, 2021. doi:
https://doi.org/10.1186/s43065-021-00021-6
6. A.E. Eiben, S.K. Smit, “Evolutionary algorithm parameters and methods to tune
them,” in Y. Hamadi, E. Monfroy, F. Saubion (eds) Autonomous search. Springer,
Berlin, Heidelberg, 2011, pp. 15–36. doi: https://doi.org/10.1007/978-3-642-21434-9
7. N. Irfan, M.C.E. Yagoub, K. Hettak, “Genetic algorithm based on efficient
tag detection in RFID reader networks,” 2011 IEEE International Conference
on Computational Intelligence for Measurement Systems and Applications
(CIMSA) Proceedings, Ottawa, ON, Canada, 2011, pp. 1–4. doi:
10.1109/CIMSA.2011.6059921
8. D. Rani, S.K. Jain, D.K. Srivastava, M. Perumal, “Genetic algorithms and their
applications to water resources systems,” in X.-S.Yang, A.H. Gandomi, S. Talatahari,
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 99
A.H. Alavi (eds) Metaheuristics in water, geotechnical and transport engineering.
Elsevier, 2013, pp. 43–78. doi: https://doi.org/10.1016/B978-0-12-398296-4.00003-9
9. A.R. Simpson, G.C. Dandy, L.J. Murphy, “Genetic algorithms compared to other
techniques for pipe optimization,” Journal of Water Resources Planning and
Management, 120(4), pp. 423–443, 1994. doi: https://doi.org/10.1061/(ASCE)0733-
9496(1994)120:4(423)
10. F. Di Pierro, S.-T. Khu, D. Savić, L. Berardi, “Efficient multi-objective optimal design
of water distribution networks on a budget of simulations using hybrid algorithms,”
Environmental Modelling & Software, 24(2), pp. 202–213, 2009. doi:
https://doi.org/10.1016/j.envsoft.2008.06.008
11. T.D. Prasad, N.S. Park, “Multiobjective genetic algorithms for design of water
distribution networks,” Journal of Water Resources Planning and
Management, 130(1), pp. 73–82, 2004. doi: https://doi.org/10.1061/(ASCE)0733-
9496(2004)130:1(73)
12. D. Kang, K. Lansey, “Revisiting optimal water-distribution system design:
Issues and a heuristic hierarchical approach,” Journal of Water Resources
Planning and Management, 138(3), pp. 208–217, 2012. doi:
https://doi.org/10.1061/(ASCE)WR.1943-5452.0000165
13. M. Cisty, Z. Bajtek, L. Celar, “A two-stage evolutionary optimization approach for an
irrigation system design,” Journal of Hydroinformatics, 19(1), pp. 115–122, 2017. doi:
https://doi.org/10.2166/hydro.2016.032
14. B. Martínez-Bahena, M.A. Cruz-Chávez, E.Y. Ávila-Melgar, M.H. Cruz-Rosales,
R. Rivera-Lopez, “Using a genetic algorithm with a mathematical programming solver
to optimize a real water distribution system,” Water, 10(10), 1318, 2018. doi:
https://doi.org/10.3390/w10101318
15. W. Bi, G. Dandy, “Optimization of water distribution systems using online retrained
metamodels,” Journal of Water Resources Planning and Management, 140(11), 2014.
doi: https://doi.org/10.1061/(ASCE)WR.1943-5452.0000419
16. S. Khalifeh, S. Akbarifard, V. Khalifeh, E. Zallaghi, “Optimization of water
distribution of network systems using the Harris Hawks optimization algorithm
(Case study: Homashahr city),” MethodsX, vol. 7, 100948, 2020. doi:
https://doi.org/10.1016/j.mex.2020.100948
17. M.S. Khorshidi, M.R. Nikoo, N. Taravatrooy, M. Sadegh, M. Al-Wardy,
G.A. Al-Rawas, “Pressure sensor placement in water distribution networks for leak
detection using a hybrid information-entropy approach,” Information Science, vol.
516, pp. 56–71, 2020. doi: https://doi.org/10.1016/j.ins.2019.12.043
18. N. Sho, A. Maki, O. Naoya, H. Jun, F. Naoki, “Monitoring of iodine species during
water purification at a public water treatment plant in Japan,” Water Supply, 19(2),
pp. 580–587, 2019. doi: https://doi.org/10.2166/ws.2018.104
19. P. Ami, O. Avi, “Multiobjective Contaminant Sensor Network Design for Water
Distribution Systems,” Journal of Water Resources Planning and Management,
134(4), pp. 366–377, 2008. doi: https://doi.org/10.1061/(ASCE)0733-
9496(2008)134:4(366)
20. Z. Hu, W. Chen, B. Chen, D. Tan, Y. Zhang, D. Shen, “Robust hierarchical sensor
optimization placement method for leak detection in water distribution systems,”
Water Resources Management, vol. 35, pp. 3995–4008, 2021. doi:
https://doi.org/10.1007/s11269-021-02922-3
21. Y. Huang, Y. Lan, S.J. Thomson, A. Fang, W.C. Hoffmann, R.E. Lacey,
“Development of soft computing and applications in agricultural and biological
Yu. P. Zaychenko, T. V. Starovoit
ISSN 1681–6048 System Research & Information Technologies, 2026, № 2 100
engineering,” Computers and Electronics in Agriculture, 71(2), pp. 107–127, 2010.
doi: https://doi.org/10.1016/j.compag.2010.01.001
22. E.H. Mamdani, “Application of fuzzy algorithms for control of simple dynamic plant,”
Proceedings of the Institution of Electrical Engineers, 121(12), 1974. doi:
https://doi.org/10.1049/piee.1974.0328
23. L. Sela, S. Amin, “Robust sensor placement for pipeline monitoring: Mixed integer
and greedy optimization,” Advanced Engineering Informatics, vol. 36, pp. 55–63,
2018. doi: https://doi.org/10.1016/j.aei.2018.02.004
24. R. Pérez et al., “Leak localization in water networks: A model-based methodology
using pressure sensors applied to a real network in Barcelona [applications of
control],” in IEEE Control Systems Magazine, vol. 34, no. 4, pp. 24–36, Aug. 2014.
doi: 10.1109/MCS.2014.2320336
25. D. Kang, K. Lansey, “Optimal Meter Placement for Water Distribution System State
Estimation,” Journal of Water Resources Planning and Management, 136(3),
pp. 337–347, 2010. doi: https://doi.org/10.1061/(ASCE)WR.1943-5452.0000037
26. I. Santos-Ruiz, F.R. López-Estrada, V. Puig, J. Blesa, “Estimation of Node Pressures
in Water Distribution Networks by Gaussian Process Regression,” 2019 4th
Conference on Control and Fault Tolerant Systems (SysTol), Casablanca, Morocco,
2019, pp. 50–55. doi: https://doi.org/10.1109/SYSTOL.2019.8864793
27. J. Mankad, B. Natarajan, B. Srinivasan, “Integrated approach for optimal sensor
placement and state estimation: A case study on water distribution networks,”
ISA Transactions, vol. 123, pp. 272–285, 2021. doi:
https://doi.org/10.1016/j.isatra.2021.06.004
28. S. Shin et al., “A systematic review of quantitative resilience measures
for water infrastructure systems,” Water, 10(2), 164, 2018. doi:
https://doi.org/10.3390/w10020164
29. D. Xu, Z. Lv, H. Zeng, H. Bessaih, B. Sun, “Sensor placement optimization in the
artificial lateral line using optimal weight analysis combining feature distance and
variance evaluation,” ISA Transactions, vol. 86, pp. 110–121, 2019. doi:
https://doi.org/10.1016/j.isatra.2018.10.039
30. H. Dong, N. Hou, Z. Wang, W. Ren, “Variance-Constrained State Estimation for
Complex Networks with Randomly Varying Topologies,” in IEEE Transactions on
Neural Networks and Learning Systems, vol. 29, no. 7, pp. 2757–2768, July 2018. doi:
https://doi.org/10.1109/TNNLS.2017.2700331
31. H. Liu, H. Yu, “Decentralized state estimation for a large-scale spatially
interconnected system,” ISA Transactions, vol. 74, pp. 67–76, 2018. doi:
https://doi.org/10.1016/j.isatra.2018.01.007
32. C. Ding, H. Peng, “Minimum redundancy feature selection from microarray gene
expression data,” Journal of Bioinformatics and Computational Biology, vol. 03, no.
02, pp. 185–205, 2005. doi: https://doi.org/10.1142/S0219720005001004
33. L.A. Zadeh, “Fuzzy sets,” Information and Control, 8(3), pp. 338–353, 1965. doi:
https://doi.org/10.1016/S0019-9958(65)90241-X
34. R. Mirabbasi, S.M. Mazloumzadeh, M.B. Rahnama, “Evaluation of irrigation water
quality using fuzzy logic,” Research Journal of Environmental Science, 2(5),
pp. 340–352, 2008. Available: https://lnk.ua/MHVvjKtE8
35. H. Gharibi, A.H. Mahvi, R. Nabizadeh, H. Arabalibeik, M. Yunesian, M.H. Sowlat,
“A novel approach in water quality assessment based on fuzzy logic,”
Journal of Environmental Management, 112, pp. 87–95, 2012. doi:
https://doi.org/10.1016/j.jenvman.2012.07.007
Hybrid computing intelligent system for assessing the stability of the water distribution system…
Системні дослідження та інформаційні технології, 2026, № 2 101
36. M. Vadiati, A. Asghari-Moghaddam, M. Nakhaei, J. Adamowski, A.H. Akbarzadeh,
“A fuzzy-logic-based decision-making approach for identification of groundwater quality
based on groundwater quality indices,” Journal of Environmental Management, 184,
pp. 255–270, 2016. doi: https://doi.org/10.1016/j.jenvman.2016.09.082
37. Soumaya Hajji et al., “Using a Mamdani Fuzzy Inference System Model (MFISM) for
Ranking Groundwater Quality in an Agri-Environmental Context: Case of the
Hammamet-Nabeul Shallow Aquifer (Tunisia),” Water, 13(18), 2507, 2021. doi:
https://doi.org/10.3390/w13182507
38. A.S. Nasr, M. Rezaei, M.D. Barmaki, “Groundwater contamination analysis using
Fuzzy Water Quality index (FWQI): Yazd province, Iran,” Geopersia, 3, pp. 47–55,
2013. Available: https://www.researchgate.net/publication/243055680_Groundwater
_contamination_analysis_using_Fuzzy_Water_Quality_index_FWQI_Yazd_provinc
e_Iran#full-text
39. Abu Rashid, Sangeeta Kumari, “Performance evaluation of ANN and ANFIS models
for estimating velocity and pressure in water distribution networks,” Water Supply,
23 (9), pp. 3925–3949, 2023. doi: https://doi.org/10.2166/ws.2023.224
40. L.A. Rossman, H. Woo, M. Tryby, F. Shang, R. Janke, T. Haxton, EPANET 2.2 User
Manual; Technical Report EPA/600/R-20/133. U.S. Environmental Protection
Agency: Washington, DC, USA, 2020.
41. L. Ferrandez-Gamot et al., “Leak localization in water distribution networks using
pressure residuals and classifiers,” IFAC-PapersOnLine, 48(21), pp. 220–225, 2015.
doi: https://doi.org/10.1016/j.ifacol.2015.09.531
42. J. Blank, K. Deb, “Pymoo: Multi-Objective Optimization in Python,” in IEEE Access,
vol. 8, pp. 89497–89509, 2020. doi: https://doi.org/10.1109/ACCESS.2020.2990567
Received 30.11.2024
INFORMATION ON THE ARTICLE
Yuriy P. Zaychenko, ORCID: 0000-0001-9662-3269, Educational and Research
Institute for Applied System Analysis of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: zaychenkoyuri@ukr.net
Tetiana V. Starovoit, ORCID: 0009-0008-6335-7679, Educational and Research
Institute for Applied System Analysis of the National Technical University of Ukraine
“Igor Sikorsky Kyiv Polytechnic Institute”, Ukraine, e-mail: starovoyt.tania@lll.kpi.ua
ГІБРИДНА ОБЧИСЛЮВАЛЬНА ІНТЕЛЕКТУАЛЬНА СИСТЕМА ДЛЯ
ОЦІНЮВАННЯ СТАБІЛЬНОСТІ СИСТЕМИ РОЗПОДІЛУ ВОДИ ТА
ВИЗНАЧЕННЯ ОПТИМАЛЬНОГО РОЗТАШУВАННЯ ДАТЧИКІВ ТИСКУ /
Ю.П. Зайченко, Т.В. Старовойт
Анотація. Подано метод визначення найкращих місць розташування датчиків
тиску в мережах водопостачання та оцінювання стану мережі за допомогою
методів штучного інтелекту. Мета – визначення вузлів мережі, які нададуть
найважливішу інформацію для виявлення витоків води та оцінювання
загального стану мережі. Вибір місць розташування датчиків ґрунтувався на
наборах даних про зміни тиску, спричинені різними сценаріями витоків,
згенерованими моделюванням EPANET. Для ранжування вузлів-кандидатів та
визначення оптимальної кількості місць розташування датчиків використано
генетичні алгоритми. Наступний крок – оцінювання стану мережі за допомогою
нейронечіткої мережі ANFIS та нейронечіткого алгоритму логічного висновку
Mamdani. Алгоритми реалізовано в середовищі Google Colab та протестовано на
ділянці мережі водопостачання в Києві, Україна.
Ключові слова: розміщення датчиків, WDN, штучний інтелект, гідравлічне
моделювання, ANFIS, Mamdani, генетичні алгоритми, EPANET 2.2.
|
| id | journaliasakpiua-article-365260 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2026-07-01T01:00:18Z |
| publishDate | 2026 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/0d/f5757d4c1dafefc804469e47c5e3830d.pdf |
| spelling | journaliasakpiua-article-3652602026-06-30T06:14:59Z Hybrid computing intelligent system for assessing the stability of the water distribution system and determining the optimum locations of pressure sensors Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску Zaychenko, Yuriy Starovoit, Tetiana розміщення датчиків WDN штучний інтелект гідравлічне моделювання ANFIS Mamdani генетичні алгоритми EPANET 2.2 sensor placement WDN artificial intelligence hydraulic modeling ANFIS Mamdani genetic algorithms EPANET 2.2 This study introduces a method for determining the best locations for pressure sensors in water supply networks and for assessing network conditions using artificial intelligence techniques. The goal is to identify the network nodes that would provide the most important information for detecting water leaks and evaluating the overall network status. The selection of sensor locations was based on data sets of pressure changes caused by various leak scenarios generated by EPANET simulations. Genetic algorithms were used to rank candidate nodes and determine the optimal number of sensor locations. The next step involved assessing the network state using the ANFIS neuro-fuzzy network and the Mamdani neuro-fuzzy logical inference algorithm. These algorithms were implemented in the Google Colab environment and tested on a section of the water supply network in Kyiv, Ukraine. Подано метод визначення найкращих місць розташування датчиків тиску в мережах водопостачання та оцінювання стану мережі за допомогою методів штучного інтелекту. Мета – визначення вузлів мережі, які нададуть найважливішу інформацію для виявлення витоків води та оцінювання загального стану мережі. Вибір місць розташування датчиків ґрунтувався на наборах даних про зміни тиску, спричинені різними сценаріями витоків, згенерованими моделюванням EPANET. Для ранжування вузлів-кандидатів та визначення оптимальної кількості місць розташування датчиків використано генетичні алгоритми. Наступний крок – оцінювання стану мережі за допомогою нейронечіткої мережі ANFIS та нейронечіткого алгоритму логічного висновку Mamdani. Алгоритми реалізовано в середовищі Google Colab та протестовано на ділянці мережі водопостачання в Києві, Україна. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2026-06-30 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/365260 10.20535/SRIT.2308-8893.2026.2.06 System research and information technologies; No. 2 (2026); 83-101 Системные исследования и информационные технологии; № 2 (2026); 83-101 Системні дослідження та інформаційні технології; № 2 (2026); 83-101 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/365260/350709 |
| spellingShingle | розміщення датчиків WDN штучний інтелект гідравлічне моделювання ANFIS Mamdani генетичні алгоритми EPANET 2.2 Zaychenko, Yuriy Starovoit, Tetiana Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску |
| title | Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску |
| title_alt | Hybrid computing intelligent system for assessing the stability of the water distribution system and determining the optimum locations of pressure sensors |
| title_full | Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску |
| title_fullStr | Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску |
| title_full_unstemmed | Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску |
| title_short | Гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску |
| title_sort | гібридна обчислювальна інтелектуальна система для оцінювання стабільності системи розподілу води та визначення оптимального розташування датчиків тиску |
| topic | розміщення датчиків WDN штучний інтелект гідравлічне моделювання ANFIS Mamdani генетичні алгоритми EPANET 2.2 |
| topic_facet | розміщення датчиків WDN штучний інтелект гідравлічне моделювання ANFIS Mamdani генетичні алгоритми EPANET 2.2 sensor placement WDN artificial intelligence hydraulic modeling ANFIS Mamdani genetic algorithms EPANET 2.2 |
| url | https://journal.iasa.kpi.ua/article/view/365260 |
| work_keys_str_mv | AT zaychenkoyuriy hybridcomputingintelligentsystemforassessingthestabilityofthewaterdistributionsystemanddeterminingtheoptimumlocationsofpressuresensors AT starovoittetiana hybridcomputingintelligentsystemforassessingthestabilityofthewaterdistributionsystemanddeterminingtheoptimumlocationsofpressuresensors AT zaychenkoyuriy gíbridnaobčislûvalʹnaíntelektualʹnasistemadlâocínûvannâstabílʹnostísistemirozpodíluvoditaviznačennâoptimalʹnogoroztašuvannâdatčikívtisku AT starovoittetiana gíbridnaobčislûvalʹnaíntelektualʹnasistemadlâocínûvannâstabílʹnostísistemirozpodíluvoditaviznačennâoptimalʹnogoroztašuvannâdatčikívtisku |