Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

Science of the Total Environment 659 (2019) 983–994

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

A simple but robust convergence trajectory controlled method for


pressure driven analysis in water distribution system
Hexiang Yan a,1, Qiongyu Wang a, Jiaying Wang a,1, Kunlun Xin a,b,⁎, Tao Tao a,b, Shuping Li a
a
College of Environmental Science and Engineering, Tongji University, Shanghai, China
b
Shanghai Institute of Pollution Control and Ecological Security, Shanghai, China

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• The proposed CTC-PDA method


provides accurate solution for PDA
problem.
• Demonstration of robustness in terms of
convergence under extreme abnormal
condition
• Self-adaptive strategy effectively
speedup the computation efficiency.
• Offer a robust hydraulic analysis tool
both for DDA and PDA
• Enrich the understanding of the impact
of relax factor on convergence behavior

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a novel convergence trajectory controlled method to perform pressure driven analysis (PDA)
Received 30 September 2018 in water distribution systems (WDSs). The proposed method makes forcibly the convergence error decrease con-
Received in revised form 11 December 2018 tinuously, which is fundamentally different from the traditional uncontrolled convergence process, thereby en-
Accepted 24 December 2018
suring a robust convergence behavior for hydraulic analysis with PDA in WDS. In addition, two Relaxation
Available online 26 December 2018
Factor section strategies are developed to control the convergence trajectory towards the desired downtrends.
Keywords:
The novel methodology is implemented based on EPANET3.0 by modifying the source code which is available
Water distribution system in GitHub (https://github.com/OpenWaterAnalytics/epanet-dev). Firstly, the improved code was validated ex-
Pressure driven analysis tensively with a benchmark WDS under rigorous boundary conditions. Subsequently, four challenging different
Convergence trajectory controlled size WDSs are also tested in terms of the effectiveness and efficiency. The results illustrate that the proposed
Pressure discharge relationship method is able to enable the convergence of PDA to be more stable and more robust, even under some extreme
EPANET3.0 abnormal boundary conditions.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction ranging from optimal design, and planning, scheduling operation, reha-
bilitation, leakage control and so on (Rossman, 2000; Todini, 2008a;
Steady state hydraulic analysis of water distribution systems Alvisi and Franchini, 2006; Giustolisi et al., 2008; Fu et al., 2012; Wang
(WDSs) is of crucial importance for solving engineering problems et al., 2014; Beh et al., 2017; Liu et al., 2017; Zheng et al., 2017; He
et al., 2018). WDS is designed and operated to deliver drinking water
with sufficient pressure and safe quantity to users. Traditionally, the
⁎ Corresponding author at: College of Environmental Science and Engineering, Tongji
University, Shanghai, China.
steady state hydraulic analysis of WDS is performed with demand
E-mail address: xkl@tongji.edu.cn (K. Xin). driven analysis (DDA) method, which implies that the pressure is
1
These authors contributed equally to this work. large enough to deliver the required water demand. During normal

https://doi.org/10.1016/j.scitotenv.2018.12.374
0048-9697/© 2018 Elsevier B.V. All rights reserved.
984 H. Yan et al. / Science of the Total Environment 659 (2019) 983–994

operating conditions, DDA method is valid for hydraulic analysis of WDS pressure and outflow are considered as the unknowns and are solved si-
since the nodal pressures are fully satisfied for the desired demand. multaneously. The convergence behavior of the Newton-Raphson
However, during pressure deficient conditions like pipe burst and method was significantly impacted due to the non-linear properties of
firefighting, the DDA method may lead to unreal solutions, which are the NPDR, thereby the convergence trajectory sometimes fluctuating
not physically available even if they are mathematically correct. The strongly, which often leads to convergence difficulty under some ex-
negative or lower pressures which are unable to deliver the desired de- treme constrain or boundary conditions. Many efforts have been made
mand are often found. To overcome this problem, the well-known pres- to achieve suitable PDA with direct consideration of NPDR. For instance,
sure driven analysis (PDA) method, which addresses the hydraulic Tanyimboh et al. (2003) proposed a PDA method with line search algo-
analysis under the pressure deficient condition, was proposed to take rithm. However, the developed model was just tested with a small case
into account the relationship between the nodal pressures and available and few conditions. Its ability for large scale WDSs has not been well
nodal outflows (Bhave, 1981). assessed. Cheung et al. (2005) modified the source code of EPANET2
Over the past few decades, many researches have been conducted to by introducing emitters into the network model to simulate pressure
develop the suitable PDA method for hydraulic analysis of WDS. The de- deficient delivered outflow using an object-oriented toolkit (OOTEN).
velopment of PDA in WDSs typically involves exploring the nodal pres- Yet this approach failed to converge when attempting to model highly
sure discharge relationship (NPDR), which is one of the basis of the looped WDS under low flow conditions. Alternatively, Giustolisi et al.
most PDA methods. The early attempt of exploring NPDR can date (2011) and Guidolin et al. (2010) developed the new solvers, WDNetXL
back to 1980s (Bhave, 1981; Germanopoulos, 1985; Salgado-Castro, and CWSNET with Excel and object-oriented C++ respectively, aiming
1988). Salgado-Castro (1988) proposed a linear relationship for NPDR. to perform hydraulic analysis under both normal and pressure deficient
Subsequently, Wagner et al. (1988) proposed a continuous quadratic conditions. More recently, Elhay et al. (2015) developed a new PDA
function to represent the partial outflow under deficient pressure. model using Goldstein's algorithm and WLS for convergence optimiza-
Then, Fujiwara and Li (1998) suggested a differentiable function, aiming tion. This approach was tested with eight challenging case study
to eliminate the numerical oscillation and convergence problem for networks.
PDA. Later, Tanyimboh and Templeman (2010) further proposed a Despite the great progress of the PDA methods in explicit or implicit
logit function based formulation to model deficient partial flow. manners, the available PDA methods are still unsatisfactory and chal-
Tabesh et al. (2014) further developed a modified NPDR based on lenged, due to the difficulty to ensure the reaching of convergence
Wagner's formula, in which both the volumetric and head dependent under extreme pressure deficient conditions. In addition, the conver-
water demand are discussed. Of these existing NPDR, Wagner's formula gence property and behavior inside the PDA solving algorithm are still
shows better match with the experimental and filed observed data, insufficiently explored and understood.
while Tanyimboh and Templeman (2010) function offers better conver- To address the issue of the convergence difficulty of iteration algo-
gence properties in PDA (Gupta and Bhave, 1996; Siew and Tanyimboh, rithm within the PDA method and the lack of knowledge on conver-
2012; Sayyed and Gupta, 2013; Shirzad et al., 2013; Vairagade et al. gence behavior within PDA, a fundamentally novel convergence
2015). controlled method is proposed in this paper. The idea of the proposed
Apart from the difficulty of defining a meaningful relationship be- convergence strategy is to directly control the convergence trajectory
tween the pressure and outflow (Ackley et al., 2001), another challenge in the iterative process, thereby ensuring the convergence of the solving
problem of PDA in WDS is to develop robust and efficient method to algorithm for PDA under both normal and extreme abnormal condi-
solve the resulting equation system. Much previous research has been tions. More precisely, the significant objectives of the present work in-
made to develop the PDA solving method. Generally, these methods volves: (i) the proposed framework of convergence trajectory
can be generally classified as implicit PDA and explicit PDA method in controlled method leads to the successive downtrend of the conver-
term of whether solving the unknown pressure and partial deliver out- gence error, and explicitly avoids the unexpected oscillation during
flow simultaneously. The implicit PDA methodology generally involves the iteration process, thereby tackling the convergence difficulty within
the traditional DDA model for hydraulics analysis and adjustment treat- PDA resulting equation system and ensuring the final convergent solu-
ment of the water demand with a repeat way, thereby making the pres- tion. It should be noted that the convergence trajectory controlled
sure and outflow consistent. For example, Todini (2008a) proposed a framework was implemented based on the EPANET3.0, which is an
PDA technique which does not involve NPDR. Baek et al. (2010) used object-oriented version of EPANET2, with some improved features and
harmony search algorithm to find the actual delivered outflow. Liu extensions, especially the PAD method (Rossman, 2010). (ii) The imple-
and Yu (2013) proposed a method called EPANET-MNO to adjust the mentation of PDA solving method using the proposed method based on
water demand. Subsequently, He et al. (2016) further improved this EPANET3.0 and extensive illustration of the robustness with different
PDA method of Liu and Yu (2013) by introducing an upper limit and a scale WDSs, including benchmark and challenging WDS cases for the
lower limit to control algorithm in the demand adjustment process, PDA problem from literature. (iii) The further exploration on knowledge
called EPANET-IMNO, thereby enhancing its robustness. regarding impact of relaxation factor on convergence behavior within
Another typical strategy of implicit PDA methodology is to use some PDA problem.
alternative artificial elements, like emitter, reservoir, valve, pipe, etc., to The remainder of the rest part of this paper is organized as follows.
equivalently represent the partial deliver outflow (Ozger and Mays, The selected NPDR equation is described briefly in next section,
2003; Todini, 2008b; Pathirana, 2010; Suribabu and Neelakantan, followed by the problem formulation of PDA of WDS and the details of
2011; Morley and Tricarico, 2014; Sayyed et al., 2014; Seyoum and the proposed methodology. Next, the proposed PDA method was tested
Tanyimboh, 2016; Mahmoud et al., 2017; Pacchin et al., 2017). Indeed, with different scale WDS cases to verify its robustness and efficiency,
these implicit PDA methods provide alternative ways to perform the hy- and then the results of these cases are presented and discussed. Finally,
draulic analysis under pressure deficient condition. However, they are a summary and conclusions are presented.
hard to implement in real time modeling and large scale real world
WDSs, especially in extended period simulation (EPS) and optimization 2. Methodology
process which generally repeats to frequently call hydraulic analysis
solver engine like EPANET2, not only due to the low efficiency by suc- Since 1980s, researches on NPDR formulas have been extensively
cessive or backtracking iterations, but also caused by the need of chang- conducted to describe the nodal outflow under all kinds of conditions
ing the topology of WDS frequently. by different users, as part of the PDA in WDSs. Almost all the researchers
The PDA method often directly introduces the NPDR into the mass agree with the point that the nodal outflow is related to the free pres-
and energy conservation equation system, where both the nodal sure head. Different types of formulas have been proposed in the
H. Yan et al. / Science of the Total Environment 659 (2019) 983–994 985

literature. Of these NPDR formulas, Wagner's formulas is the most Then the GGA can be written as:
widely used one (Wagner et al., 1988), which is fit well with experi- 8  −1  
mental data since it is closely similar to the orifice outflow formula < Hk ¼ A D −1 A A21 Q k−1 þ q−A21 D11 −1 A12 Q k−1 −A21 D11 −1 A10 H0
21 11 12
form (Shirzad et al., 2013). And the formula is a continuous function,  
: Q k ¼ Q k−1 −D −1 A Q k−1 þ A Hk þ A H
11 11 12 10 0
with upper and lower bounds consistent with WDS's real behavior. It
is expressed as: ð5Þ
8
> 0; H ≤Hmin where k is the iteration trials.
>
> !1=n i i
>
< For the pressure driven analysis, namely under the pressure defi-
Hi −H min
qi ¼ qi req i
; Hmin bH i bH req ð1Þ cient condition, part of the nodal required demand cannot be satisfied
>
> H req min i i
>
> i −H i due to the drop of the pressure. If the PDA model is taken into consider-
:
qreq
i ; Hreq
i bH i ation, Eq. (2) can be rewritten as:
    
where q∗i is the available discharge at node i, qreq
i is the required demand A11 A12 Q −A10 Η0
¼ ð6Þ
at node i, Hi is the available pressure at node i, Hmin
i is the minimum pres- A21 0 H −q
sure (for nodal pressure lower than this, the available discharge is zero),
Hreq
i is the minimum required pressure head (for nodal head greater where q* is the vector of actual flow of nodes, which is described by the
than this, the required demand can be satisfied), and n is the head expo- NPDR in the current paper (Wagner et al., 1988), as shown in Eq. (1).
nent which is considered between 1.5 and 2. The derivation of Eq. (6) gives:
    
2.1. Pressure driven analysis (PDA) model D11 A12 dQ dE
¼  ð7Þ
A21 D22 dH dq
As previously stated, the present study builds based on EPANET3.0
(Rossman, 2010), an object-oriented version the new updated where
EPANET2.0, and released as open source package in GitHub (https://
github.com/OpenWaterAnalytics/epanet-dev) under MIT license, in dq
D22 ¼ ð8Þ
which some features have been revised and some new features have dH
been added. More importantly, the current EPANET3.0 not only includes 
dq ¼ A21 Q þ q ð9Þ
the revised DDA model, but that also supports the pressure driven anal-
ysis (PDA) model with directly introduction of the NPDR in the resulting
Submitting the PDA into GGA gives:
equation system. Both of the DDA and the PDA in EPANET3.0 are solved
using the global gradient algorithm (GGA) framework, which is recog- 8  −1  
< Hk ¼ A D −1 A −D A21 Q k−1 þ q −A21 D11 −1 A12 Q k−1 −A21 D11 −1 A10 H0
nized the most efficient algorithm for steady state hydraulic analysis 21 11 12 22
 
: Q k ¼ Q k−1 −D −1 A Q k−1 þ A Hk þ A H
in WDS. For sake of better interpretation of the proposed PDA method, 11 11 12 10 0

the PDA equation system based on GGA is briefly described for ð10Þ
reminder.
The objective of GGA is to solve the continuity and energy equations Solving Eq. (10) gives the temporary Hk and Qk. Then, ΔHk−1 and
of WDS, given as (Todini and Pilati, 1988): ΔQk−1 can be calculated as:
    
A11 A12 Q −A10 Η0
¼ ð2Þ ΔHk−1 ¼ Hk −Hk−1 ð11Þ
A21 0 H −q
ΔQ k−1 ¼ Q k −Q k−1

where A12 ¼ A21 T is the topological incidence matrix (np × nn) that de- The iteration form of Eq. (11) is
fines the pipe and node connectivity with the unknown nodal head, A10
¼ A01 T is the topological incidence matrix (np × no) that defines the Hk ¼ Hk−1 þ λΔHk−1 ð12Þ
pipe and node connectivity with the known nodal head, QT = [Q1, Q k ¼ Q k−1 þ λΔQ k−1
Q2, ……, Qnp] is the pipe flow vector (1 × np), qT = [q1, q2, ……, qnn] is
the water demand vector (1 × nn), HT = [H1, H2, ……, Hnn] is the un- where λ is the relaxation factor and k is the iteration trials.
known pressure head vector (1 × nn), H0T = [H01, H02, ……, H0no] is
the known head vector (1 × no), nn and no are the number of unknown 2.2. Proposed convergence trajectory controlled pressure driven analysis
and known pressure head nodes, respectively, np is the links' number (CTC-PDA) method
with unknown flow rate. A11 represents the diagonal matrix. More de-
tails can refer to Todini and Pilati (1988). As stated previously, the proposed method here is developed and
Eq. (2) can be differentiated with respect to the pipe flow rates and implemented based on EPANET3.0, within which the well-known clas-
nodal heads to give: sical GGA is used to solve the PDA problem. Generally, the target vari-
ables H and Q in Eq. (12) are updated by iteration, and the relaxation
    
D11 A12 dQ dE factor λ is equal to 1 in EPANET2 for DDA. However, as pointed out by
¼ ð3Þ Todini (2008a), the PDA problem trends to reduce the convergence of
A21 0 dH dq
the original algorithm due to the introduction of the NPDR, which signif-
where icant impacts the convergence behavior during the iteration process. λ
= 1 is no longer always suitable for all the iterative step, which means
that sometimes the relaxation factor of 1 may lead to divergence. There-
dQ ¼ Q k −Q k−1 fore, the PDA problem is transformed to find a suitable λ to ensure the
dH ¼ Hk −Hk−1 ð4Þ convergence. This is generally done by defining the convergence itera-
dE ¼ A11 Q k−1 þ A12 Hk þ A10 H0
tive error as the objective function, thereby searching a suitable λ for
dq ¼ A21 Q k−1 þ q
the current step iteration with optimization algorithm.
986 H. Yan et al. / Science of the Total Environment 659 (2019) 983–994

The basic idea of the proposed method in this paper is to forcibly where ε is the convergence tracker, and Δh and Δq are given as:
make the convergence error decreasing by choosing suitable λ. Theoret-
ically, convergent solution can be reached if a suitable λ is found at each Δh ¼ HfromX
−Hto −hfrom‐to
ð20Þ
iteration step. Within the proposed methodology, a convergence error Δq ¼ q þ Q
formula, which is used as alternative process controlling tracker (called
convergence tracker hereafter), is defined. Two different relaxation fac- ||Δh||2 and ||Δq||2 indicate the correctness of the continuity and en-
tor searching strategies are developed to find a feasible λ value, which ergy equations of WDS. The convergence direction can be controlled
makes the next convergence tracker less than the current one. Different with the defined convergence tracker by ensuring εk b εk−1. Four con-
from the optimization method such as Elhay et al. (2015), in which the vergence criteria, as expressed in Eq. (21), are adapted to judge conver-
computation expensive Jacobi and Hessen matrix construction is gence of the calculation. The PDA solving iteration calculation would
needed, a simple uniformly enumerated relaxation factor sampling ap- stop when all the four convergence criteria are satisfied simultaneously,
proach with specific step size Δλ is developed to find the suitable relax- or the maximum iteration steps are reached. It should be note that the
ation factor. In order to ensure to find out a suitable λ, a recursive convergence tracker ε and the criterion 3 φ∞(Δhk), criterion 4 φ∞(Δqk)
searching process, which enlarges the density of the sample, is per- are different both in definition and function even though they are all re-
formed if no feasible λ is found in the current enumerated sample lated to the iterative balance errors of continuity and energy. The former
value set. The details of the proposed method are described below. is a combination of 2-norm continuity and energy iterative errors,
which is used to observe dynamically the convergence behavior and
2.2.1. Convergence tracker to control the convergence process. While the latter two are the ∞-
Before defining convergence tracker formula, some basic definitions norm of the closure errors of energy and continuity equations respec-
and convergence criteria are predefined. tively, which are stricter and used to judge the convergence.
A 2-norm of a vector x = (x1, x2,⋯, xk)T is defined as: h        iT
Error ¼ φE qk ; φ∞ Q k ; φ∞ Δh ; φ∞ Δqk
k
vffiffiffiffiffiffiffiffiffiffiffiffiffi ; Tolerance
u k
T
uX
kxk2 ¼ t x2i ð13Þ ¼ εq ; εQ ; εΔh ; ε Δq ð21Þ
i¼1
where εq, εQ, εΔh and εΔq are the threshold of the four convergence
T
A ∞-norm of a vector x = (x1, x2,⋯, xk) is defined as: criteria, respectively.

kxk∞ ¼ max jxi j ð14Þ 2.2.2. Convergence trajectory controlled strategy


1bibk
As shown in Fig. 1, two relaxation factor searching strategies were
Criterion 1, the total change rate of total nodal demand during twice developed to control the convergence direction during the iteration
continuous iterations is less than a given threshold value: process. The detailed operation structures of the two strategies are illus-
trated in Fig. 1.
  Pqk −qk−1  Regarding the first convergence controlling strategy (Fig. 1(a)), the
φE q k ¼ P k  ≤εq ð15Þ following steps at kth iteration are carried out to determine a suitable
q
relaxation factor λ:
Criterion 2, the ∞-norm of changing of pipe flows during twice con- 1. Enumerating a set of λ in the entire candidate interval (0, 1] with a
tinuous iterations is less than a given threshold value: specific initial step size Δλ.
    2. Calculating the convergence tracker ε for each λ in the generated λ
 
φ∞ Q k ¼ Q k −Q k−1  ≤ε Q ð16Þ sample set.

3. Finding the minimum εkmin by sort calculation.
Criterion 3, the ∞-norm of the closure error of pipe energy during 4. Checking: if εkmin b εk−1, then λ which corresponds to εkmin is chosen
each iteration is less than a given threshold value (i.e., conservation of as the suitable relaxation factor for the current trial; otherwise εkmin ≥
energy): εk−1, decreasing the step size Δλ and going back to step 1.
5. Checking the global convergence tolerance (four convergence
   
  criteria): stopping or performing the next iteration.
φ∞ Δh ¼ Hkfrom −Hkto −hfrom‐to  ≤εΔh
k k
ð17Þ

Within the this strategy, since the λ, which is chosen as the suitable
Criterion 4, the ∞-norm of the closure error of continuity equation is relaxation factor, corresponds to the minimum convergence tracker
less than a given threshold value: value εkmin in the numeration sample set, this strategy is named as
    best ever relaxation factor (for short as ‘brf’ hereafter) searching strategy.
 
φ∞ Δqk ¼ qk þ ∑Q k  ≤εΔq ð18Þ A simple unified enumeration approach with a specific step size Δλ is

adopted to generate λ sample set, coupling with a circulation adjusting
The changes during twice continuous iterations are considered in process if the required λ is not found. For example, initially Δλ = 0.2 is
Criterion 1 and Criterion 2, which ensure the convergence mathemati- chosen as looping step size. If Δλ = 0.2 fails to satisfy εkmin b εk−1, then
cally. Criterion 3 and Criterion 4 assess the closure errors of energy and Δλ will be decreased to Δλ = 0.1 as looping step size in a self-adaptive
continuity equations, which ensure the correctness of hydraulic calcula- process. Then if εkmin b εk−1, then the algorithm goes to the next step;
tion, and prevent the calculation from pseudo convergence. Criterion 3 otherwise, Δλ will be decreased again to continue to enlarge the enu-
and Criterion 4 are physically based. If both of Criterion 3 and Criterion meration samples of λ set. This self-adaptive adjusting treatment can
4 are satisfied, Criterion 1 and Criterion 2 must be reached. surely enhance the ability to find out the suitable relaxation factor and
For sake of well controlling and tracking the convergence process, a is called recursive tracking.
convergence tracker, which combines the 2-norm continuity iterative In consideration of the computational demand for calculation of con-
error and energy iterative error, is defined and introduced to track and vergence tracker function of the λ samples set in the whole interval (0,
control the convergence behavior, given as: 1], another strategy, called adaptive relaxation factor (for short as ‘arf’
hereafter) searching strategy, is also proposed. In this strategy, for
ε ¼ kΔhk2 þ kΔqk2 ð19Þ each iteration, the convergence tracker εk corresponding to λ = 1 in
H. Yan et al. / Science of the Total Environment 659 (2019) 983–994 987

Fig. 1. The two proposed convergence trajectory controlled strategies. (a) The best ever relaxation factor searching strategy, (b) the adaptive relaxation factor searching strategy.

Eq. (19) is calculated firstly, if εk b εk−1 is already satisfied, then the al- 2.3. Study cases
gorithm goes to the next step directly; otherwise, the (0, 1] interval enu-
meration approach is implemented, and the recursive tracking To clearly assess and demonstrate the performance of the proposed
treatment is still performed if necessary. CTC-PDA methodology, it has been extensively tested, verified, and il-
The following steps are carried out by adaptive relaxation factor lustrated not only by a benchmark WDS case from literature with a
strategy. broad series of boundary conditions, ranging from normal scenarios to
extreme abnormal conditions, but also by four challenging WDS cases
1. Calculating εk with λ = 1, checking: if εk b εk−1, then going to step 6. in different size and complexity, which are available in the public
Otherwise, going to step 2. ASCE library (Elhay et al., 2015).
The first WDS case N1 is originally taken from Tabesh (1998) and
2. Enumerating a set of λ in the entire candidate interval (0, 1] with a
later used by a number of researchers as a benchmark to validate their
specific initial step size Δλ.
PDA methods. This network consists of a fixed head reservoir feeding
3. Calculating the convergence tracker εk for each λ in the generated
8 demand nodes through 12 pipes. The normal demands of other
set.
nodes (node 2–8 shown in Fig. 2) are all set as 20.8 L/s, the same as
4. Finding the minimum εkmin by sort calculation,
other researchers for the sake of performance comparison
5. Checking: if εkmin b εk−1, the λ that corresponds to εkmin is chosen as
(Tanyimboh et al., 2003; Tabesh et al., 2014). The fixed pressure head
the suitable relaxation factor in the current trial; if εkmin ≥ εk−1, de-
of the reservoir is equal to 100 m in the current analysis. For simplicity,
creasing the step size Δλ and going to step 1.
all the elevation of nodes are set as 0 m. A pressure deficient scenario is
6. Checking the global convergence tolerance (four convergence
introduced through a large required demand at node 9 in the network
criteria): stopping or performing the next iteration.
system, as shown in Fig. 2. A broad series of large required demands
ranging from normal conditions to extreme abnormal conditions were
The detailed algorithm of the two proposed strategies was incor- tested with the proposed CTC-PDA method. The configurations of re-
porated to the EPANET3.0 by modifying the original source code. The lated parameters of N1 are shown in Table 1. More details can also
corresponding names of proposed CTC-PDA method coupling with refer to Tanyimboh et al. (2003) and Tabesh et al. (2014), in which
these two strategies are called for short as CTC-PDAbrf and CTC- the same example network was adopted for PDA method test.
PDAarf, respectively. These modified versions EPANET3.0 codes Solving practical engineering problem is the most initial motivation
were tested with five representative WDSs, which are demonstrated for the PDA topic, therefore, except for the small typical benchmark
in next section. WDS, it is of significant importance to assess the performance of the
988 H. Yan et al. / Science of the Total Environment 659 (2019) 983–994

among which the maximum one is about 30 times of the smaller de-
mand (q = 10 L/s) and accounts for about 67% of the total demand for
this benchmark network. It should be pointed out that the extreme
water demands configuration at node 9 is to evaluate the robustness
of the PDA method in the mathematical consideration, even though
this is unlikely to happen in reality. Preliminary, all the different de-
mand conditions at node 9 were tested with the traditional DDA
method by EPANET2. The simulation results by DDA method illustrate
that all the water demand conditions give the unreal solutions (unrea-
sonable negative pressure) except for the cases of demand q = 10 L/s
and q = 20 L/s, which confirm that all the water demand scenarios, ex-
cept for the cases q = 10 L/s and q = 20 L/s, result in pressure deficient
conditions. The results by EPANET2 have not been given in this paper
due to the space limitation. For the purpose of robustness evaluation,
the simulation results are compared between the proposed CTC-PDA
method with different strategies and the original PDA (called as ORI-
PDA hereafter) solver within the current EPANET3.0 as well as the
other PDA methods available in literature.
Fig. 2. Network layout of example WDS case N1 (Tabesh, 1998). To firstly verify the correctness of the proposed CTC-PDA method, a
required demand of 62.5 L/s at node 9 for N1, which is the same config-
proposed CTC-PDA method on WDSs with different size and complex- uration to the other researches, is performed, and the corresponding re-
ity, which are equivalent to the real world WDSs. To achieve this goal, sults as well as the results from reference literature are summarized in
four challenging WDSs of different size were tested with abnormal sce- Table 3. In comparing the pressure and available outflow, the proposed
narios in the present work. The four selected example WDSs are from CTC-PDA method provides the exact same results to that by Tanyimboh
Elhay et al. (2015) for PDA method evaluation. The layout of the four et al. (2003), Pathirana (2010) and Tabesh et al. (2014), and the similar
WDSs is shown in Fig. 3. And the general information of these WDSs is results with Tabesh (1998) and CWSNET. This suggests that the pro-
summarized in Table 2. Following the setting in Elhay et al. (2015), posed method is able to give the good results as the available PDA
the nodal demands of the four example WDSs cases N2–N5 were mag- method with the same accuracy.
nified by multiplying them by a factor of five, thereby ensuring that the Subsequently, Table 4 shows the simulation results of the hydraulic
problem was actually a PDA problem, rather than a DDA problem. analysis under different normal and abnormal demand conditions of N1
All calculations were performed on a computer with an Intel proces- both by the original PDA method within EPANET3.0 (ORI-PDA) and the
sor, Core i5-4570 CPU with 3.2 GHz, and 64-bit Windows 7. It should be proposed CTC-PDA method with two different relaxation factor
noted that even though the CTC-PDA method was realized based on searching approaches. As can be observed in Table 4, the proposed
EPANET3.0 with GGA in the current work, the CTC-PDA method is ge- CTC-PDA method provides reasonable solutions for all the different
neric, which means it can be applied to the similar iterative solving algo- large demand conditions, in which the actual pressures and available
rithm for steady state hydraulic analysis of WDS. outflows are consistent in term of the selected NPDR. This confirms
that the proposed CTC-PDA method works well even with extreme ab-
3. Results and discussions normal water demand configurations (at node 9), including the maxi-
mum one which accounts for N60% of the total demand.
The performance of the proposed CTC-PDA method was demon- Comparatively, the ORI-PDA gives an incorrect solution after maximum
strated both in effectiveness and efficiency with the example WDSs. iteration trials (1000 steps) under an extreme large water demand (q
The term CTC-PDA without the subscript ‘brf’ and ‘arf’ denotes that = 300 L/s, see the results with bold italics Table 4). This implies that
both of the two strategies are included for results presentation. the proposed CTC-PDA method shows higher robustness comparing
with ORI-PDA in convergence capability under extreme abnormal defi-
3.1. Convergence ability of CTC-PDA method cient pressure conditions.
Fig. 4 illustrates the convergence iteration process under different
Unlike some previous researches, in which only one or a few high demand configurations with both the ORI-PDA and the CTC-PDA
water demand conditions were tested for performance assessment, in method. It can be seen that the ORI-PDA is able to achieve the conver-
order to extensively assess the robustness of the proposed CTC-PDA gence in a few iterations for all the demand configurations considered
method, a series of normal and abnormal water demand conditions (de- at node 9, except for 270 L/s, 290 L/s, and 300 L/s, in which the conver-
mands ranging from q = 10 L/s to q = 300 L/s with an interval of Δq = gence trajectory oscillated strongly and is non-convergent after maxi-
10 L/s) at a specific node 9 for this benchmark network N1 were tested, mum iteration trials (1000 steps). On the contrary, the proposed CTC-
PDA method converges rapidly and stably for all the demand configura-
Table 1 tions, which shows better robustness in convergence ability. Fig. 5 dem-
Configuration diameter D, roughness Cw, length L, elevation E, minimum pressure Hmin, onstrates the observed convergence tracker (error) with respect to the
required pressure Hreq, water demand q for example WDS N1. iteration trials for both the ORI-PDA and the CTC-PDA method with ini-
Pipes information Nodes information
tial demand of 270 L/s and 280 L/s, respectively. Fig. 5(a) illustrates the
convergence trajectory of CTC-PDAbrf and oscillating non-convergence
Pipe D Cw L Node E Hmin Hreq q (L/s)
trajectory of the ORI-PDA till reaching the maximum iterative step,
identifier (mm) (m) identifier (m) (m) (m)
and Fig. 5(b) shows both convergence trajectory by CTC-PDAbrf and
1-2, 1-4 250 130 1000 1 0 0 30 –
ORI-PDA. In Fig. 5(a), the solid green line indicates the controlled con-
(reservoir)
2-3, 4-7 175 130 1000 2, 4 0 0 30 20.8 vergence trajectory for CTC-PDAbrf, with which the convergence tracker
2-5, 4-5 145 130 1000 3, 7 0 0 30 20.8 rapidly reaches the desired tolerance in 8 iteration trials, while the
3-6, 7-8 115 130 1000 5 0 0 30 20.8 dashed red line denotes uncontrolled iterative trajectory for ORI-PDA,
5-6, 5-8 100 130 1000 6, 8 0 0 30 20.8 which fluctuates strongly (details shown in Fig. 6), and fails to converge
6-9, 8-9 100 130 1000 9 0 0 30 10–300
after maximum steps for the case of water demand 270 L/s.
H. Yan et al. / Science of the Total Environment 659 (2019) 983–994 989

Fig. 3. Network layout of example WDSs N2–N5.

Furthermore, in terms of water demand of 280 L/s, although both of the In spite of the better robustness for searching convergence of the
ORI-PDA and the CTC-PDA method converged finally, the CTC-PDA proposed CTC-PDA when comparing with the ORI-PDA under some ex-
method enables a more stable and faster convergence trajectory (see treme rigorous conditions, the inside convergence searching property is
Fig. 5(b)). still not clear and well understood. In order to further understand the
On the whole, both of the ORI-PDA and the proposed CTC-PDA convergence searching behavior, the typical relaxation factor λ during
method show good performance for iteration convergence ability, the iterative processes of ORI-PDA and CTC-PDA were traced and ex-
while the latter one outperforms the former for some extreme rigorous tracted for the typical demand of 270 L/s at node 9 in WDS N1, as
demand conditions for WDS N1. shown in Table 5. In original EPANET3.0, the relaxation factor is set as
λ = 1 if the convergence tracker is larger than 1. Otherwise λ is set as
Table 2 0.5 or 1, which can give a smaller convergence tracker in the next itera-
Basic properties summary of example WDSs of N2–N5. tive step. Correspondingly, the proposed method compulsively searches
Network identifier nodes information Pipes information an ever best relaxation factor, which ensures the next convergence
tracker to be smaller than the current one. It is supposed that a confi-
Nnode qtotal (L/s) Npipe Dmax (mm) Ltotal (km)
dential relaxation factor interval which leads to convergence exists.
N2 934 144.82 848 1000 75.596
Consequently, a suitable relaxation factor for each local iterative step
N2 1976 1924.35 1770 600 116.895
N4 2465 5220.73 1890 1000 593.889 is vital for convergence searching direction and convergence speed.
N5 14,830 2478.67 12,523 1200 1844.547 From Table 5, in the first 8 trials, the CTC-PDA method can find some
Note: Nnode and Npipe denote the number of the nodes and pipes respectively, qtotal and Ltotal
suitable relaxation factor and rapidly reach the convergence. On the
denote the total demand and the total length of the WDS, and Dmax denotes the maximum contrary, the ORI-PDA always sets relaxation factor as 1, which may be
diameter of pipe in the WDS. beyond the confidential interval for convergence, thereby leading to
990 H. Yan et al. / Science of the Total Environment 659 (2019) 983–994

Table 3
The results by the proposed CTC-PDA method and the other PDA methods available in literature with required demand q = 62.5 L/s at node 9 for example WDS N1.

Node ID Tanyimboh et al., Tabesh, 1998 Tabesh et al., 2014 Pathirana, 2010 CWSNET Proposed CTC-PDA
2003

H (m) qout (L/s) H (m) qout (L/s) H (m) qout (L/s) H (m) qout (L/s) H (m) qout (L/s) H (m) qout (L/s)

1 100 171.8 100 171.1 100 171.8 100 171.8 100 173.34 100 171.8
2, 4 88.2 −20.8 88.02 −20.8 88.21 −20.8 88.21 −20.8 85.77 −20.8 88.21 −20.8
3, 7 71.36 −20.8 70.94 −20.8 71.38 −20.8 71.38 −20.8 68.07 −20.8 71.38 −20.8
5 71.99 −20.8 71.58 −20.8 72.01 −20.8 72 −20.8 69.08 −20.8 72 −20.8
6, 8 36.68 −20.8 35.33 −20.8 36.73 −20.8 36.71 −20.8 39.81 −20.8 36.71 −20.8
9 5.27 −26.2 4.28 −25.5 5.28 −26.2 5.27 −26.2 5.76 −27.74 5.27 −26.2

Note: the minus ‘−’ in water demand column qout denotes the available outflow in the node of WDS. The results of ‘CWSNET’ are from Tabesh et al. (2014).

continuous fluctuating of convergence trajectory (see Fig. 6) or slow the nodes, which is the same as Elhay et al. (2015), for introducing pres-
and rough convergence trajectory (see Fig. 5(b)). This could explain sure deficient conditions. For comparison analysis purpose, the results
why the CTC-PDA method shows better robustness of convergence by original EPANET3.0 and available in literature were used. The statis-
under some extreme rigorous water demand configurations. tical results are shown in Table 6, where the pressures are classified ac-
For extensively evaluating the robustness of the CTC-PDA method, cording to the NPDR with Hmin and Hreq (Hmin = 0 m, Hreq = 30 m for all
four different benchmark networks equivalent to real world WDSs cases).
with different sizes were also tested. The water demands for the four According to the tested results (see Table 6), both of the ORI-PDA
tested WDSs were all set as five times of the original demand for all and the proposed CTC-PDA methods are able to achieve the final

Table 4
The results by original PDA method in EPANET3.0 and the proposed CTC-PDA method with different demand conditions for WDS N1.

Node 9 demand (L/s) ORI-PDA CTC-PDAbrf & arf (with Δλ = 0.01)

Actual pressure H (m) Node 9 available outflow (L/s) Actual pressure H (m) Node 9 available outflow (L/s)

Node Node Node Node Node Node Node Node Node Node

2&4 3&7 5 6&8 9 2&4 3&7 5 6&8 9

10 90.19 76.88 77.76 56.17 50.9 −10 90.19 76.88 77.76 56.17 50.9 −10
20 89.08 73.8 74.55 44.53 27.78 −19.25 90.19 73.8 74.55 45.53 27.78 −19.25
40 88.43 71.99 72.65 38.95 11.24 −24.48 88.43 71.99 72.65 38.95 11.24 −24.48
60 88.23 71.42 72.05 36.86 5.67 −26.09 88.23 71.42 72.05 36.86 5.67 −26.09
80 88.14 71.19 71.8 35.99 3.35 −26.75 88.14 71.19 71.8 35.99 3.35 −26.75
100 88.1 71.07 71.68 35.57 2.2 −27.07 88.1 71.07 71.68 35.57 2.2 −27.07
120 88.08 71.01 71.61 35.33 1.55 −27.25 88.08 71.01 71.61 35.33 1.55 −27.25
140 88.07 70.97 71.57 35.18 1.15 −27.35 88.07 70.97 71.57 35.18 1.15 −27.35
160 88.06 70.94 71.54 35.08 0.88 −27.43 88.06 70.94 71.54 35.08 0.88 −27.43
180 88.05 70.92 71.52 35.02 0.7 −27.48 88.05 70.92 71.52 35.02 0.7 −27.48
200 88.05 70.91 71.51 34.97 0.57 −27.52 88.05 70.91 71.51 34.97 0.57 −27.52
220 88.04 70.9 71.5 34.93 0.47 −27.54 88.04 70.9 71.5 34.93 0.47 −27.54
240 88.04 70.89 71.49 34.91 0.4 −27.57 88.04 70.89 71.49 34.91 0.4 −27.57
260 88.04 70.89 71.49 34.89 0.34 −27.6 88.04 70.89 71.49 34.89 0.34 −27.6
280 88.04 70.88 71.48 34.87 0.29 −27.58 88.04 70.88 71.48 34.87 0.29 −27.58
300 128.26 219.21 166.46 696.97 2832.31 −3.23 88.04 70.88 71.48 34.86 0.25 −27.6

The unconverged results provided by ORI-PDA with demand q = 300 L/s were marked with bold-italics.

Fig. 4. The iteration process of PDA with different demands in N1.


H. Yan et al. / Science of the Total Environment 659 (2019) 983–994 991

Fig. 5. Iteration convergence tracker with demands of 270 L/s and 280 L/s at node 9 of N1.

Fig. 6. Iteration convergence tracker by ORI-PDA with demand of 270 L/s at node 9 of N1.

Table 5
Relaxation factors in iteration process for WDS N1 with demand of 270 L/s at node 9 by corresponding available outflows are zero. This is theoretically correct
different PDA methods. according to NPDR when the available pressure is less than the Hmin.
Iterative step Relaxation factor
Therefore, the solutions for the four cases by the CTC-PDA are correct
and acceptable. This suggests that the proposed method is able to
CTC-PDAbrf (Δλ = 0.01) ORI-PDA
achieve the convergence solution effectively and offer a realistic result
1 0.29 1 for large scale and complicated WDSs. In addition, what is noteworthy
2 1 1
is that the number of the negative pressure provided by CTC-PDA is sig-
3 0.13 1
4 1 1 nificant smaller than that provided by Elhay et al. (2015). Based on the
5 0.05 1 current tested results, it is hard to infer the reasonable explanation for
6 1 1 this difference since the initial conditions, NPDR parameters are not ex-
7 1 1 actly the same for these cases.
8 1 1
Result Convergent Non-convergent
3.2. Performance for different relaxation factor searching strategies and
computing efficiency
convergence for all example cases of N2–N5, and give the almost the
same results. According to the results from N2 and N3 by the ORI-PDA In spite of the good performance demonstrated extensively in the
and the CTC-PDA, as well as the statistical results of N2–N5 from Elhay previous statement, another important concert for the PDA method is
et al. (2015), negative pressures are observed, with which the the computation demand, which may lead to shortage of computer

Table 6
The number of the pressure classification by different PDA method.

WDS identifier H ≥ 30 0 ≤ H b 30 Hb0

Elhay et al., 2015 ORI-PDA CTC-PDA Elhay et al., 2015 ORI-PDA CTC-PDA Elhay et al., 2015 ORI-PDA CTC-PDA

N2 – 636 636 – 198 198 50 14 14


N3 – 1442 1442 – 323 323 34 5 5
N4 – 1724 1724 – 166 166 21 0 0
N5 – 11,461 11,461 – 1062 1062 85 0 0
992 H. Yan et al. / Science of the Total Environment 659 (2019) 983–994

Table 7
Convergence iteration trails Niter for CTC-PDAarf and CTC-PDAbrf strategies.

Step size Δλ N2 N3 N4 N5

CTC-PDAbrf CTC-PDAarf CTC-PDAbrf CTC-PDAarf CTC-PDAbrf CTC-PDAarf CTC-PDAbrf CTC-PDAarf

0.01 14 16 16 17 17 16 17 19
0.02 14 16 19 17 18 17 19 19
0.05 14 16 19 17 16 17 19 22
0.1 14 16 18 17 17 16 21 23
0.2 15 16 17 17 21 17 19 23
0.5 15 16 17 17 21 17 23 23

Table 8
Convergence computation time Titer (s) for CTC-PDAarf and CTC-PDAbrf method.

Step size Δλ N2 N3 N4 N5

CTC-PDAbrf CTC-PDAarf CTC-PDAbrf CTC-PDAarf CTC-PDAbrf CTC-PDAarf CTC-PDAbrf CTC-PDAarf

0.01 1.87 0.11 4.68 0.49 6.20 1.27 43.03 4.08


0.02 0.98 0.09 2.80 0.36 3.26 0.76 22.19 2.90
0.05 0.53 0.10 1.39 0.26 1.38 0.43 10.59 2.43
0.1 0.32 0.09 0.76 0.29 1.06 0.34 7.10 1.98
0.2 0.20 0.10 0.46 0.22 0.76 0.33 4.19 1.90
0.5 0.16 0.10 0.34 0.21 0.32 0.22 2.82 1.74

resource for some practical or research problems like WDS optimal de- general, the convergence iteration trials arise with the increasing of
sign problem, which needs to recall the hydraulic solver frequently. the step size Δλ, while the convergence computation demands decrease
As stated previously in Section 2, two different strategies, namely the with the increasing of the step size Δλ. In terms of computation effi-
CTC-PDAbrf and CTC-PDAarf were proposed to search a suitable relaxa- ciency, the CTC-PDAarf outperforms the CTC-PDAbrf according to the re-
tion factor at each local iteration trial, which enables the method to sults of different size WDSs.
make the convergence trajectory in a downtrend, thereby ensuring Fig. 7 demonstrates varying trend of the GGA convergence iteration
the final convergent solution. Considering the similarity to the real trials and convergence computation demand with respect to the step
world WDS, the four cases N2–N5 were also used to analyze the CTC- size of relaxation factor in case N5 (the largest size and most complex
PDA with the best relaxation factor (brf) searching strategy and adap- case among the five example WDSs). From Fig. 7(a), the convergence it-
tive relaxation factor (arf) searching strategy. eration steps, in general, shows a slightly raising trend with the increase
In terms of the CTC-PDAbrf, the specific step size Δλ, which is used to of step size of relaxation factor both for the CTC-PDAbrf and CTC-PDAarf.
generate an enumerated set of relaxation factor, is set as 0.01, 0.02, 0.05, This implies that much effort spending on finding a better relaxation
0.1, 0.2, and 0.5. And then the corresponding set of relaxation factor is just offers little convergence speedup benefit. On the contrary, increas-
calculated as {1Δλ, 2Δλ, …, nΔλ}, with 0 b nΔλ ≤ 1. In terms of the ing of the step size of relaxation factor would significantly reduce the
CTC-PDAarf, the unique difference comparing to the CTC-PDAbrf is to computation demand, especially for CTC-PDAbrf. This finding demon-
check the next convergence tracker εk with λ = 1 firstly. If εkbεk−1, strates the importance of initial step size in terms of computation effi-
the λ = 1 is selected to update Hk and Qk, otherwise, the λ selection is ciency optimization for CTC-PDA. That is, the initial step size of Δλ =
determined by the same way as the brf selection process. Table 7 0.5 may be suggested for the purpose of computational efficiency bene-
shows the total iteration steps used to search the final convergent solu- fit. In addition, as it can be observed in Fig. 7(b) that the computation
tion for the four WDSs, and Table 8 illustrates the corresponding com- consumption by CTC-PDAarf is less than CTC-PDAbrf, the smaller step
putation demand used to searching the final convergent solution. In size, the more obvious benefit of computation resource saving. The

Fig. 7. Computation performance versus step size for the proposed relaxation factor searching strategy. (a) Convergence trials under different step size with the proposed two relaxation
factor strategies, (b) convergent time consuming under different step size with the proposed two relaxation factor strategies.
H. Yan et al. / Science of the Total Environment 659 (2019) 983–994 993

Table 9 (i) The proposed CTC-PDA method could effectively and efficiently
Convergence performance by different PDA methods for example cases of N2–N5. perform the steady state hydraulic analysis for different size
WDS identifier ORI-PDA CTC-PDAbrf CTC-PDAarf WDSs both under normal and abnormal pressure deficient con-
(Δλ = 0.5) (Δλ = 0.5) ditions. This implies that the proposed convergence trajectory
Niter Titer (s) Niter Titer (s) Niter Titer (s) controlled strategy seems superior to the traditional uncon-
trolled convergence strategy in term of convergent stability and
N2 16 0.098 15 0.16 16 0.1
N3 16 0.188 17 0.34 17 0.21 robustness, in particular for extreme rigorous pressure deficient
N4 15 0.212 21 0.32 17 0.22 conditions.
N5 27 2.245 23 2.82 23 1.74 (ii) The proposed two damping relaxation factor searching strategies
best ever relaxation factor and adaptive relaxation factor are able to
well implement the convergence trajectory controlled frame-
fact that the CTC-PDAarf outperforms the CTC-PDAbrf in computation ef- work and provide stable and accurate results. This suggests that
ficiency implies that the benefit from reduction of enumeration of relax- the enumerative searching approach is able to be an alternative
ation factor is superior to the benefit from a better relaxation factor at simple way to find the feasible relaxation factor within the
each local iterative trial. convergence trajectory controlled framework. In addition, the
Table 9 demonstrates the overall convergence performance both in adaptive relaxation factor searching strategy shows better perfor-
convergence iteration trials and computation times by ORI-PDA and mance than the best ever relaxation factor searching strategy with
CTC-PDA with step size Δλ = 0.5, which is the most efficient one respect to computational efficiency. This suggests that the relax-
among the tested values. The proposed CTC-PDA method shows com- ation factor plays an importance role for convergence of PDA
parative ability comparing to the ORI-PDA with respect to convergence problem.
iteration trials. Meanwhile, in terms of computation efficiency, the CTC- (iii) In comparing the overall performance between the proposed
PDAarf is also comparative to the ORI-PDA, while the CTC-PDAarf is infe- CTC-PDA and the original PDA (ORI-PDA) in EPANET3.0, the for-
rior to the ORI-PDA according to the example cases of N2–N5. As a mer outperforms the latter in terms of the convergence stability
whole, the ORI-PDA with EPANET3.0 shows a good performance on and robustness, especially for some extreme pressure deficient
PDA problem both for small and large size of WDSs, except for some ex- conditions; whereas the former is inferior to or comparative to
treme rigorous pressure deficient conditions. On the contrary, the pro- the latter in computational efficiency.
posed CTC-PDA demonstrates a more stable and robust convergence
capability and comparative computational efficiency only with adaptive
In closing, the outcome of this paper not only has provided an alter-
relaxation factor searching strategy (step size Δλ = 0.5) when compar-
native PDA method that can perform the hydraulic analysis of WDSs ef-
ing with ORI-PDA. The CTC-PDA coupling with best ever relaxation factor
fectively based on EPANET3, but also has built fundamental knowledge
searching strategy is inferior to ORI-PDA in terms of computational effi-
regarding the impacts of damping relaxation factor on convergence pro-
ciency due to the high demand of convergence error function calcula-
cess within the PDA. In addition, the resulting tool of this work can be
tion. Further optimization is needed in suitable relaxation factor
applied to practical PDA problem in WDS such as optimal design, leak-
searching under the CTC-PDA methodology for efficiency improvement.
age detection modeling, and so on.

4. Summary and conclusions Acknowledgement

Steady state hydraulic analysis of water distribution systems This work was supported by the National Natural Science Founda-
(WDSs) under pressure deficient condition is significant important in tion of China [Grant No. 51808396 and 51678425], and Major Science
urban water supply system design and management, especially for the and Technology Program for Water Pollution Control and Treatment
abnormal conditions such pipe burst, firefighting, leakage detection, in [2017ZX07201003].
which the traditional demand driven analysis (DDA) method often
fails. This consequently leads to many studies to develop robust and ef- References
ficient pressure driven analysis (PDA) method to perform the correct
Ackley, J.R.L., Tanyimboh, T.T., Taher, B., Templeman, A.B., 2001. Head-driven analysis of
hydraulic analysis. However, their convergence performances of state- water distribution systems. Computing and Control for the Water Industry, CCWI
of-art PDA algorithms available in literature have been still unsatisfac- 2001, Leicester, U.K. 1, pp. 183–192.
tory when dealing with some extreme abnormal required demand con- Alvisi, S., Franchini, M., 2006. Near-optimal rehabilitation scheduling of water distribution
systems based on a multi-objective genetic algorithm. Civ. Eng. Environ. Syst. 23 (3),
ditions even in a small scale WDS. This is because the convergence 143–160.
properties of the resulting equation system has been weaken due to Baek, C.W., Jun, H.D., Kim, J.H., 2010. Development of a PDA model for water distribution
the introduction of the non-linear nodal pressure discharge relationship systems using harmony search algorithm. KSCE J. Civ. Eng. 14, 613–625.
Beh, E.H., Zheng, F., Dandy, G.C., Maier, H.R., Kapelan, Z., 2017. Robust optimization of
(NPDR) under the pressure deficient condition. water infrastructure planning under deep uncertainty using metamodels. Environ.
This paper has proposed a simple but robust convergence trajectory Model. Softw. 93, 92–105.
controlled pressure driven analysis (CTC-PDA) methodology that ex- Bhave, P.R., 1981. Node flow analysis of water distribution systems. J. Transp. Eng. 107 (4),
457–467.
plicitly accounts for convergence process. In the proposed methodology, Cheung, P., Van Zyl, J.E., Reis, L.F.R., 2005. Extension of EPANET for pressure driven de-
the convergence direction has been controlled in a downtrend iterative mand modeling in water distribution system. Computing and Control in the Water
trajectory, which ensures the final convergence of the PDA solving algo- Industry CCWI 2005, Exeter, U.K. 1, pp. 311–316.
Elhay, S., Piller, O., Deuerlein, J., Simpson, A., 2015. A robust, rapidly convergent method
rithm. In addition, a physical concerted convergence tracker, which
that solves the water distribution equations for pressure-dependent models.
strictly ensures the mass and energy balance within the specific toler- J. Water Resour. Plan. Manag. 142 (2), 04015047.
ance, has been defined. Two different damping relaxation factor Fu, G., Kapelan, Z., Kasprzyk, J., Reed, P., 2012. Optimal design of water distribution sys-
tems using many-objective visual analytics. J. Water Resour. Plan. Manag. 139 (6),
searching strategies, which explicitly control the convergence trajec-
624–633.
tory, have been developed. Five different size benchmark case studies Fujiwara, O., Li, J., 1998. Reliability analysis of water distribution networks in consider-
have been performed under both normal and abnormal outflow condi- ation of equity, redistribution, and pressure-dependent demand. Water Resour. Res.
tions to demonstrate the effectiveness of the proposed CTC-PDA 34, 1843–1850.
Germanopoulos, G., 1985. A technical note on the inclusion of pressure dependent de-
methods. Based on the results, several conclusions can be outlined mand and leakage terms in water supply network models. J. Civ. Eng. Syst. 2 (3),
below: 171–179.
994 H. Yan et al. / Science of the Total Environment 659 (2019) 983–994

Guidolin, M., Burovskiy, P., Kapelan, Z., Savic, D.A., 2010. CWSNet: an object-oriented Sayyed, M.A.H., Gupta, R., 2013. Predicting deficient condition performance of water dis-
toolkit for water distribution system simulations. Proceedings of the 12th Interna- tribution networks. Civ. Eng. Infrastruct. J. 46 (2), 161–173.
tional Conference on Water distribution system analysis conference, WDSA2010, Sayyed, M.A.H., Gupta, R., Tanyimboh, T.T., 2014. Modelling pressure deficient water dis-
12–15 September 2010, Tucson, AZ, U.S, pp. 1–13. tribution networks in EPANET. Proc. Eng. 89, 626–631.
Giustolisi, O., Savic, D.A., Kapelan, Z., 2008. Pressure-driven demand and leakage simula- Seyoum, A.G., Tanyimboh, T.T., 2016. Investigation into the pressure-driven extension of
tion for water distribution networks. J. Hydraul. Eng. 134 (5), 626–635. the EPANET hydraulic simulation model for water distribution systems. Water
Giustolisi, O., Savic, D., Berardi, L., Laucelli, D.J., 2011. An Excel-based solution to bring Resour. Manag. 30, 5351–5367.
water distribution network analysis closer to users. Computing and Control in the Shirzad, A., Tabesh, M., Farmani, R., Mohammadi, M., 2013. Pressure-discharge relations
Water Industry, CCWI 2005, Exeter, U.K. 3, pp. 805–810. with application to head-driven simulation of water distribution networks. J. Water
Gupta, R., Bhave, P.R., 1996. Comparison of methods for predicting deficient-network per- Resour. Plan. Manag. 139, 660–670.
formance. J. Water Resour. Plan. Manag. 122, 214–217. Siew, C., Tanyimboh, T.T., 2012. Pressure-dependent EPANET extension. Water Resour.
He, P., Tao, T., Xin, K.L., Li, S.P., Yan, H.X., 2016. Modelling water distribution systems with Manag. 26, 1477–1498.
deficient pressure: an improved iterative methodology. Water Resour. Manag. 30, Suribabu, C.R., Neelakantan, T.R., 2011. Balancing reservoir based approach for solution to
593–606. pressure deficient water distribution networks. Int. J. Civ. Struct. Eng. 2, 639–647.
He, G.L., Zhang, T.Q., Zheng, F.F., Zhang, Q.Z., 2018. An efficient multi-objective optimiza- Tabesh, M., 1998. Implications of the Pressure Dependency of Outflows on Data Manage-
tion method for water quality sensor placement within water distribution systems ment, Mathematical Modeling and Reliability Assessment of Water Distribution Sys-
considering contamination probability variations. Water Res. 143, 165–175. tems. (PhD Thesis). University of Liverpool, Liverpool, U.K.
Liu, J., Yu, G.P., 2013. Iterative methodology of pressure-dependent demand based on Tabesh, M., Shirzad, A., Arefkhani, V., Mani, A., 2014. A comparative study between mod-
EPANET for pressure-deficient water distribution analysis. J. Water Resour. Plan. ified and available demand driven based models for head driven analysis of water
Manag. 139, 34–44. distribution networks. Urban Water J. 11 (3), 221–230.
Liu, H., Savić, D.A., Kapelan, Z., Creaco, E., Yuan, Y., 2017. Reliability surrogate measures for Tanyimboh, T.T., Templeman, A.B., 2010. Seamless pressure-deficient water distribution
water distribution system design: comparative analysis. J. Water Resour. Plan. Manag. system model. Proc. Inst. Civ. Eng. Water Manag. 163, 389–396.
143 (2), 04016072. Tanyimboh, T.T., Tahar, B., Templeman, A.B., 2003. Pressure-driven modelling of water
Mahmoud, H.A., Savic, D., Kapelan, Z., 2017. New pressure-driven approach for modeling distribution systems. Water Sci. Technol. Water Supply 3 (1–2), 255–261.
water distribution networks. J. Water Resour. Plan. Manag. 143 (8), 04017031. Todini, E., 2008a. On the convergence properties of the different pipe network algorithms.
Morley, M.S., Tricarico, C., 2014. Pressure Driven Demand Extension for EPANET Water Distribution Systems Analysis Symposium 2006, pp. 1–16.
(EPANETpdd). Technical Report. University of Exeter, Exeter, U.K, p. 2008. Todini, E., 2008b. Towards realistic extended period simulations (EPS) in looped pipe net-
Ozger, S.S., Mays, L.W., 2003. A semi-pressure-driven approach to reliability assessment of work. Water Distribution Systems Analysis Symposium 2006, pp. 1–16.
water distribution networks. Proceeding, 30th International Association of Hydraulic Todini, E., Pilati, S., 1988. A gradient algorithm for the analysis of pipe networks. Wiley,
Research Congress, Theme B-Urban and Rural Water Systems for Sustainable Devel- London, pp. 1–20.
opment. Aristoteleio Panepistimio Thessalonikis, Thessaloniki, Greece, pp. 345–352. Vairagade, S.A., Sayyed, M.A., Gupta, R., 2015. Node head flow relationships in skeleton-
Pacchin, E., Alvisi, S., Franchini, M., 2017. Analysis of non-iterative methods and proposal ized water distribution networks for predicting performance under deficient condi-
of a new one for pressure-driven snapshot simulations with EPANET. Water Resour. tions. World Environmental and Water Resources Congress 2015, pp. 810–819.
Manag. 31 (1), 75–91. Wagner, J.M., Shamir, U., Marks, D.H., 1988. Water distribution reliability - simulation
Pathirana, A., 2010. EPANet2 desktop application for pressure driven demand modeling. methods. J. Water Resour. Plan. Manag. 114, 276–294.
In: Lansey, K.E., Choi, C.Y., Ostfeld, A., Pepper, I.L. (Eds.), Proceedings of the 12th Inter- Wang, Q., Guidolisi, M., Savic, D., Kapelan, Z., 2014. Two-objective design of benchmark
national Conference on Water distribution system analysis conference, WDSA2010, problems of a water distribution system via MOEAs: towards the best-known ap-
12–15 September 2010, Tucson, AZ, pp. 65–74. proximation of the true Pareto front. J. Water Resour. Plan. Manag. 141, 04014060.
Rossman, L.A., 2000. EPANET 2: User's Manual. U.S. Environmental Protection Agency, Zheng, F.F., Zecchin, A.C., Newman, J.P., Maier, H.R., Dandy, G.C., 2017. An adaptive
Cincinnati. convergence-trajectory controlled ant colony optimization algorithm with applica-
Rossman, L.A., 2010. An overview of EPANET version 3.0. Proceedings of the 12th Interna- tion to water distribution system design problems. IEEE Trans. Evol. Comput. 21,
tional Conference on Water Distribution System Analysis Conference, WDSA2010, 773–791.
12–15 September 2010, Tucson, AZ, U.S, pp. 14–18.
Salgado-Castro, R.O., 1988. Computer Modelling of Water Supply Distribution Networks
Using the Gradient Method. (Ph.D. thesis). Univ. of Newcastle, Newcastle upon
Tyne, U.K.

You might also like