Beruflich Dokumente
Kultur Dokumente
Research note
KEYWORDS Abstract The fundamental goal of a pipeline utility is to serve its customers with a low cost water supply
Optimization; of acceptable quality. The number, type, size, and location of transient protective devices play a direct role
Genetic algorithm (GA); in the pipeline system reliability and expenditure. The purpose of this study is to optimize the design
Fuzzy inference system of these devices to prevent water column separation after source pump power failure. The minimum
(FIS); pressures along the pipeline are assumed to be higher than 10 m to avoid water column separation. A
Transient flow; rational, systematic, and efficient optimization algorithm is constructed by combining the Fuzzy Inference
Protective devices. System (FIS) and the Genetic Algorithm (GA). The FIS representing expert knowledge is incorporated into
the GA approach to improve its fitness evaluation process. Three cases are presented to demonstrate the
effectiveness and efficiency of the proposed hybrid approach.
2013 Sharif University of Technology. Production and hosting by Elsevier B.V.
Open access under CC BY license.
maintained using a genetic algorithm. The GA deals with a large Mamdani method is intuitive and has widespread acceptance,
number of discrete or continuous variables, does not require while the Sugeno method is computationally efficient and
a derivable objective function, explores a wide search space works well with optimization and adaptive techniques. The
simultaneously, provides a population of optimum solutions, Sugeno method also has guaranteed continuity of the output
and works with numerically generated data, experimental data, surface [17]. To demonstrate the effectiveness of the proposed
or analytical functions in extremely complex problems [7]. In hybrid approach, three cases are presented.
this paper, GA is improved using fuzzy inference systems. A
fuzzy decision making is incorporated in the GA approach to 2. Optimization algorithm
improve its fitness evaluation process and its capability for
handling constraints. The fitness evaluation in this paper does The optimization algorithm is combined with a transient
not incorporate cost directly. Each chromosome is evaluated simulation program to achieve an optimal solution. In ordinary
using a fuzzy decision defined after transient analysis. genetic algorithms, the fitness function that evaluates each
The basic idea underlying fuzzy logic was suggested by chromosome derives from a mathematical formula describing
Zadeh [8]. In general, fuzzy logic is concerned with formal the objective function. Definition of a fitness function is an
principles of approximate reasoning, while classical two-valued important and difficult task. The fitness evaluation in this paper
logic (true or false) is concerned with formal principles of does not incorporate cost directly. In the present work, a fuzzy
reasoning. Fuzzy logic uses the continuum of logical values decision making is incorporated in the GA approach to improve
between 0 (completely false) and 1 (completely true). Two its fitness evaluation process and its capability of handling
of the main concepts that play an important role in many constraints. The objective of the proposed method is to optimize
applications of fuzzy logic are the concepts of linguistic variable one-way surge tanks with a least cost design subject to the
and fuzzy if-then rules [9]. For example, height is a linguistic following constraints:
variable when its values are defined to be low, medium or
high. Each linguistic value is represented as a fuzzy set that is 1. Minimum pressures along the pipeline should be higher
characterized by a membership function, usually taking values than 10 m to avoid water column separation.
between 0 and 1. In general, a fuzzy rule can be represented as: 2. Final liquid height in the tanks should be sufficient to
prevent vortices at the surge tank outlet.
If x1 is A1 and x2 is A2 and . . . xn is An then
As will be explained in the next section (Section 3),
y1 is B1 and y2 is B2 and . . . ym is Bm, maintaining the desirable final liquid height in the surge
where x1 , x2 , . . . , xn , y1 , y2 , . . . , ym are linguistic variables, and tanks leads to the least cost design of these devices. The final
A1 , A2 , . . . , An , B1 , B2 , . . . , Bm are their respective linguistic liquid height constraint is regarded as the objective of the
values. The goal of using fuzzy systems is to put human optimization subject to the minimum head constraint. In this
knowledge into engineering systems in a systematic, efficient, method, each feasible chromosome (solution) that satisfies
and analyzable order. Fuzzy systems are knowledge-based or the pressure constraint after transient analysis is evaluated
rule-based systems and work very well for many engineering in the range of [0, 1] using a fuzzy inference system, with
problems [10]. respect to the final liquid height in the surge tanks, as shown
Goulter and Bouchart [11] used fuzzy sets combined in Figure 1. The fitness value for infeasible chromosomes or
with linear programming for network cost minimization. chromosomes that violate the minimum pressure constraint is
Vamvakeridou-Lyroudia [12] used fuzzy sets for pressure and set to zero without FIS interference. Normalizing the fitness
velocity constraint violation in a dynamic programming algo- value in the range [0, 1], it will be possible to find the
rithm, for optimal design of water supply networks. Xu and global optimum solution. Using the fuzzy inference system,
Goulter [13] presented a fuzzy linear program optimization determination of the fitness value for each chromosome and
method in which the capital costs of the network were min- satisfying the constraints, are accomplished in an easy and
imized while maintaining the nodal heads at demand nodes transparent manner. A generalized method is obtained for
within a satisfactory region, as defined by the customers evaluation of the fitness value that does not change with
at those nodes. Revelli and Ridolfi [14] simulated uncertain dimensional characteristics, time, or location.
parameters, like the roughness coefficient of pipes and the Mamdani and Sugeno are two types of fuzzy inference
demands of the network, using fuzzy theory. Vamvakeridou- system that are tested and compared in this paper. In these
Lyroudia et al. [15] used a fuzzy multi-objective optimization fuzzy systems, minimum is used as the AND operator, and
model (minimizing cost and maximizing a benefit-quality func- maximum is used as the OR operator. The weighted average
tion) to the Anytown water distribution network. They used and the centroid methods are used as the defuzzification
genetic algorithms, combined with fuzzy reasoning, for benefit- methods in Sugeno and Mamdani FIS methods, respectively.
quality evaluation. They showed that their model manages to The input variables and the rules of these two fuzzy systems are
find a better solution than any other previous approach in terms the same. In Mamdani-type inference, the output membership
of cost, despite the multiple criteria applied for the benefit functions are fuzzy sets, but Sugeno output membership
function being more extensive and stricter. Amirabdollahian functions are constant. The binary genetic algorithm has been
et al. [16] applied a fuzzy genetic algorithm to obtain the least- written in MATLAB, version 7.6.0.324 (R2008a). Tournament
cost design of looped water distribution networks. They used selection and elitism (retaining the best solution at each
a fuzzy decision system to eliminate the traditional use of the generation) are used. The probability of uniform crossover is
penalty function in the genetic algorithm. They concluded that 0.8 and the mutation probability is 0.05. The algorithm stops
their proposed method yielded solutions with reduced costs. when the fitness value reaches one or when the number of
Mamdani and Sugeno are two types of fuzzy inference generations is sufficient to converge. The appropriate number
system that are tested and compared in this paper. In Mamdani- of generations will be noted in each case. The optimization
type inference, the output membership functions are fuzzy sets, algorithm is flexible enough to deal with a variety of pipelines,
but Sugeno output membership functions are constant. The changing the system characteristics and membership functions.
280 M.R. Chamani et al. / Scientia Iranica, Transactions A: Civil Engineering 20 (2013) 278285
into a fuzzy system. When transient analysis is completed, 5. Results and discussions
the final liquid height in each surge tank may be insufficient
(less than 1.0 m and unable to prevent bathtub vortices), 5.1. Case 1
sufficient (about 1.0 m), or excessive (more than 1.0 m). Each
input comprises three membership functions (low, good, and As already stated, column separation occurs at about 5.197 s
high), if it is insufficient, sufficient, or excessive, respectively, after complete pump power failure. In this case, optimization of
as presented in Figure 3. The horizontal axis represents the three one-way surge tanks in a system including an air chamber
final water surface height and the vertical axis represents is considered to maintain allowable pressures. The air chamber
membership values for low, good and high. The upper is located at the pump station and the initial air volume is set at
and lower bounds depend on the size of the surge tanks and 2.8 m3 (100 ft3 ). Since a one-way surge tank can be placed on
are determined by the optimization algorithm. If the liquid each node, a total number of 42 series pipes, starting with pipe
height in the tank reaches zero and the pressure at the tank number one at the upstream, are considered instead of the three
connection drops to below zero, the liquid height in the surge existing pipes (three segments in the shortest pipe, 23 and 16
tank will be presented by a negative value, as shown in the segments in the two other pipes). Four different sizes (1, 1.25,
low membership function in Figure 3. This means that the 1.5, and 1.75 m) are considered for the diameter of each surge
size of the one-way surge tank must be increased. The inputs tank. The ratio of initial liquid height to diameter in each surge
are considered to be the minimum, maximum, and mean values tank is assumed to be 1.0 from which the height of each tank can
among the final liquid height in the surge tanks. The output is be calculated after its diameter is determined. The diameter of
the fitness value in the range of [0, 1], as shown in Figure 4. The the connection pipe to the pipeline and the discharge coefficient
fuzzy system is constructed based on 10 rules. The infeasible in the surge tanks are assumed to be 0.3 m and 0.8, respectively.
rules are omitted. The location and size (diameter) of three one-way surge
tanks are the design variables. Therefore, there are 4,093,440
4. Description of pipeline system (41 40 39 4 4 4) possible combinations in the
complete enumeration for this case. The population size is
The pipeline used for this study is that studied by 100 and the algorithm stops when the fitness value reaches
Watters [18]. A transient analysis program is used to determine one or when there are 50 generations. The fuzzy system
the pressure head and velocity along the pipeline system after used is a three-input-one-output system with 10 rules. The
pump power failure. The code of this program is presented in three inputs are the minimum, maximum, and mean values
FORTRAN [3], which is converted to metric units. It is based among the final liquid height in the tanks, and the output is
on the method of characteristics. The system includes a pump the fitness value. The result of the optimization shows that
station at the upstream and a reservoir at the downstream, as only two one-way surge tanks will be sufficient to overcome
shown in Figure 5. The pump station has four similar (source) column separation. The initial liquid height in one of the three
pumps in parallel to deliver 694 lit/s from an elevation of surge tanks does not decrease during transient simulation and
120.4 m to an elevation of 256 m. Pipeline specifications are two other tanks are sufficient to prevent column separation.
presented in Table 1. Each pump has five stages and the rotary Therefore, this tank is omitted and the simulation program
moment of inertia for each pump and motor unit is 20 kg m2 . runs again with two remaining tanks. Mamdani and Sugeno
The minimum pressures along the pipeline are assumed to results are presented in Table 2. It is shown that both Mamdani
be higher than 10 m to avoid water column separation, and Sugeno methods yield the same results, but the Sugeno
based on absolute atmospheric pressure, fluid temperature, system shows a slower convergence in this case. The numbers
and vapor pressure. The real time of transient simulation after of generations to convergence in Mamdani and Sugeno types
power failure is set at 100 s. The results of the analysis show are 5 and 13, respectively, as shown in Figure 6. Although
that column separation occurs at about 5.197 s after complete both methods achieve the same solutions, a slight difference is
pump power failure. A software package HYTRAN [22] is used observed between the two maximum fitness values in the two
Pipe no. Upstream Downstream Length (m) Diameter (m) Material Friction factor Wave
elevation (m) elevation (m) speed (m/s)
FIS method Optimization results of one-way surge tanks Transient analysis results Maximum Number of
fitness value generations
No. Optimal Diameter (m) Initial Final Minimum Maximum Consumed
locationa liquid liquid head (m) air water
height (m) height (m) volume (m3 ) (m3 )
Alternative no. No. of surge tanks Locationa Tank diameter (m) Initial liquid height (m) Minimum head (m)
FIS method Optimization results of one-way surge tanks Transient analysis results Maximum Number of
fitness value generations
No. Optimal Diameter (m) Initial liquid Final liquid Minimum Consumed
locationa height (m) height (m) head (m) water (m3 )
1 30 2.00 2.00
Mamdani 7.00 0.03 14.94 0.50 13
2 36 2.50 2.50
1 30 2.00 2.00
Sugeno 7.00 0.03 14.94 0.50 13
2 36 2.50 2.50
a
The number refers to the pipe number where a one-way surge tank is located at downstream end of the pipe.
not change with dimensional characteristics, time (implying [5] Stephenson, D. Simple guide for design of air vessels for water hammer
inflation and discount rates in costs), and location. The stopping protection of pumping lines, J. Hydraul. Eng., 128(8), pp. 792797 (2002).
[6] Izquierdo, J., Lpez, P.A., Lpez, G., Martnez, F.J. and Prez, R. Encapsu-
criteria of the optimization algorithm become rational, and local lation of air vessel design in a neural network, Appl. Math. Model., 30,
optimum solutions are avoided by normalizing the fitness value pp. 395405 (2006).
in the range [0, 1]. The number, size, and location of the surge [7] Haupt, R.L. and Haupt, S.E., Practical Genetic Algorithms, 2nd Edn., John
Wiley & Sons Inc., Hoboken, New Jersey (2004).
tanks have no significant effects on the volume of consumed
[8] Zadeh, L.A. Fuzzy sets, Inf. Control, 8(3), pp. 338353 (1965).
water to overcome water column separation. In other words, [9] Zadeh, L.A. Fuzzy logic, IEEE Comput., 21(4), pp. 8393 (1988).
the amount of water consumed by one-way surge tanks to [10] Wang, L.X., A Course in Fuzzy Systems and Control, Prentice-Hall, Inc., New
effectively handle water column separation is approximately Jersey (1997).
[11] Goulter, I.C. and Bouchart, F. Fuzzy programming in water distribution
constant and does not change with changes in surge tank network design, In Proc. of 1st Int. Conf. on Comput. Methods and Water
geometry or location. Mamdani and Sugeno methods yield Res., Rabat, pp. 3344, Springer, Morocco, Berlin (1988).
similar results for the demonstrated cases. Fulfillment of the [12] Vamvakeridou-Lyroudia, L.S. Fuzzy reasoning in water supply network
optimization, In Proc. of the 4th Int. Conf. on Application of Artif. Intell. to Civ.
minimum pressure constraint and critical submergence shows
and Struct. Engrg., pp. 255262, Civil-Comp Press, Cambridge, Edinburgh,
the effectiveness of the fuzzy-genetic method proposed. UK (1995).
[13] Xu, C. and Goulter, I.C. Optimal design of water distribution networks
using fuzzy optimization, Civ. Engrg. Environ. Syst., 16, pp. 243266
References (1999).
[14] Revelli, R. and Ridolfi, L. Fuzzy approach for analysis of pipe networks, J.
[1] Jung, B.S. and Karney, B.W. Hydraulic optimization of transient protection Hydraul. Eng., 128(1), pp. 93101 (2002).
devices using GA and PSO approaches, J. Water Resour. Plann. Manage., [15] Vamvakeridou-Lyroudia, L.S., Walters, G.A. and Savic, D.A. Fuzzy multi-
132(1), pp. 4452 (2006). objective optimization of water distribution networks, J. Water Resour.
[2] Jung, B.S., Karney, B.W., Boulos, P.F. and Wood, D.J. The need for Plann. Manage., 131(6), pp. 467476 (2005).
comprehensive transient analysis of distribution systems, J. Am. Water [16] Amirabdollahian, M., Chamani, M.R. and Asghari, K. Optimal design of
Works Assn., 99(1), pp. 112123 (2007). water networks using fuzzy genetic algorithm, Proc. ICEWater Manag.,
[3] Larock, B.E., Jeppson, R.W. and Watters, G.Z., Hydraulics of Pipeline Systems, 164(7), pp. 335346 (2011).
CRC Press, New York (2000). [17] Fuzzy Logic Toolbox Users Guide, The MathWorks, Inc. (2008).
[4] Vtkovsk, J.P., Simpson, A.R. and Lambert, M.F. Leak detection and [18] Watters, G.Z., Analysis and Control of Unsteady Flow in Pipelines, 2nd Edn.,
calibration of water distribution systems using transients and genetic Butterworth Publishers, Boston, Massachusett (1984).
algorithms, Water Dist. Syst. Conf., Div. of Water Resour. Plan. ASCE, Tempe, [19] Yldrm, N. and Kocaba, F. Critical submergence for intakes in open
Arizona, pp. 5764 (June 79, 1999). channel flow, J. Hydraul. Eng., 121(12), pp. 900905 (1995).
M.R. Chamani et al. / Scientia Iranica, Transactions A: Civil Engineering 20 (2013) 278285 285
[20] Yldrm, N. and Kocaba, F. Critical submergence for intakes in still-water Shokoufeh Pourshahabi received a B.S. degree in Civil Engineering, in
reservoir, J. Hydraul. Eng., 124(1), pp. 103104 (1998). 2002, from Shahid Bahonar University of Kerman, Iran, and an M.S. degree
[21] Yldrm, N., Kocaba, F. and Glcan, S.C. Flow-boundary effects on in Civil Engineering (Water Resources Engineering), in 2008, from Isfahan
critical submergence of intake pipe, J. Hydraul. Eng., 126(4), pp. 288297 University of Technology, Isfahan, Iran. Since 2009, she has been lecturer in the
(2000). Department of Civil Engineering at the Institute of Higher Education of Eqbal
[22] Lawgun, N. Water hammer software: hytran solutions, Auckland, New Lahoori, Mashhad, Iran. Her research interests include: hydraulic transients,
Zealand. Available in: http://www.hytran.net. evolutionary algorithms, fuzzy inference systems, artificial neural networks,
and extreme values.
Mohammad Reza Chamani received a B.S. degree in Civil Engineering, in Farid Sheikholesalm received a B.S. degree in Electronics from Sharif
1985, from Amirkabir University of Technology, Tehran, Iran, and M.S. and University of Technology, Tehran, Iran in 1990, and M.S. and Ph.D. degrees
Ph.D. degrees in Water Resources Engineering from the University of Alberta, in Communications and in Electrical Engineering from Isfahan University
Edmonton, Canada, in 1993 and 1997, respectively. He joined the Department of Technology, Iran, in 1994 and 1998, respectively. Since 1999, he has
of Civil Engineering at Isfahan University of Technology, Iran, in 1998, and his been with the Department of Electrical and Computer Engineering at Isfahan
current position is Associate Professor. His research interests include, but are University of Technology, Iran, where he is currently Associate Professor
not limited to hydraulic structures, pipe flow, and river engineering. He is also of Electrical Engineering. His research interests include: control algorithms,
author of several books and papers. stability analysis, nonlinear systems, intelligent control and robotics.