Evolutionary Algorithms for the Determination of Critical

Depths in Conduits
A. Kanani1; M. Bakhtiari2; S. M. Borghei3; and D.-S. Jeng4
Downloaded from ascelibrary.org by University of Southern Queensland on 05/11/16. Copyright ASCE. For personal use only; all rights reserved.

Abstract: The determination of critical depth and the position of control sections is important in open-channel hydraulics. Calculation of
critical depth in open channels is useful not only for determining the condition of a flow but also for hydraulic design and analysis of
experimental and analytical results. In this study, unlike the conventional approaches, an alternative method, based on a genetic algorithm
共GA兲, for the calculation of critical depth in conduits is presented. In our model, the governing equations are transferred into an objective
function that is then minimized using a GA in order to calculate critical depth. This method does not have the limitations of existing
empirical and semiempirical methods and can be used for any prismatic or nonprismatic open-channel cross section. The concepts
presented in this paper can be generalized for solving other tortuous hydraulic engineering equations and problems.
DOI: 10.1061/共ASCE兲0733-9437共2008兲134:6共847兲
CE Database subject headings: Open channels; Algorithms; Conduits; Hydraulic design.

Introduction presented a computational method to determine the critical depth

and control points in open channels. Straub 共1978兲 developed
The accurate determination of critical depths and the location of semiempirical relations for the calculation of critical depth for
control sections are important parameters in open-channel hy- several common channel shapes 共e.g., rectangular, trapezoidal,
draulic design. The critical depth concept is used as the basis for circular and elliptical兲. All equations proposed in Straub 共1978兲
flow characterization, i.e., sub- or supercritical, while the deter- were limited to a specific range. Chaudhry and Ballamudi 共1988兲
mination of control sections is essential for the calculation of presented a computational procedure for the prediction of critical
water surface profiles in gradually varied flow problems. Numer- depths in general symmetrical compound channels. They incorpo-
ous analytical approximations for the calculation of critical depth rated a general definition of the Froude number based on the
in open channels of a simple rectangular or triangular cross sec- one-dimensional unsteady momentum and continuity equations.
tion have been used 共French 1986兲. For channels with more- Wang 共1998兲 presented an explicit analytical solution for the
complicated shapes 共e.g., a trapezoidal cross section兲, numerical equation of critical depth in a trapezoidal channel, which is valid
methods such as the Newton–Raphson method have to be used to over a large range of parameters and its maximum error is limited
solve the equations. No doubt, for even more irregular cross sec- to within 0.014. Kurbanov and Khanov 共2004兲 developed an ana-
tions, such as an egg-shaped, horse shoe, or a natural channel, lytical solution for the calculation of critical depth that is valid
numerical simulation will become increasingly difficult and time only for symmetrical–polygonal cross sections. Later, Swamee
consuming. and Rathie 共2005兲 developed exact analytical solutions for critical
Since the 1970’s, considerable efforts have been devoted to depth in trapezoidal open-channel sections, based on a form of
finding a general and accurate method for prediction of critical fast converging infinite series.
depth and its position 共Smith 1972; Straub 1978; Chaudhry and All aforementioned investigations have been limited and cus-
Ballamudi 1988; Wang 1998; Dey 2001; Kurbanov and Kurbanov tomized for a specific type of cross section. For example, the
2004; Swamee and Rathie 2005兲. Among these, Smith 共1972兲 methods of Wang 共1998兲 and Swamee and Rathie 共2005兲 are valid
only for trapezoidal cross sections and the method of Kurbanov
1
and Khanov 共2004兲 is valid only for symmetrical–polygonal cross
Research Associate, Dept. of Civil Engineering, Sharif Univ. of Tech- sections. Thus, a more-generalized model for prediction of critical
nology, P.O. Box 11365-8639, Tehran, Iran. E-mail: akanani@alum.
depth in open channels is desired. In this paper, a general model
sharif.edu
2
Research Associate, Dept. of Civil Engineering, Sharif Univ. of Tech-
with a genetic algorithm 共GA兲 for calculation of critical depth
nology, P.O. Box 11365-8639, Tehran, Iran. E-mail: m_bakhtiari@ will be presented. The advantages of the proposed model are high
alum.sharif.edu accuracy, efficiency, and applicability over a range of cross
3 sections.
Professor, Dept. of Civil Engineering, Sharif Univ. of Technology,
P.O. Box 11365-8639, Tehran, Iran. E-mail: [email protected]
4
Professor, Division of Civil Engineering, School of Engineering,
Physics and Mathematics, Univ. of Dundee, Dundee DD1 4HN, Scotland, Evolutionary Algorithms
U.K. 共corresponding author兲. E-mail: [email protected]
Note. Discussion open until May 1, 2009. Separate discussions must
be submitted for individual papers. The manuscript for this paper was Soft computing methods are a collection of new computational
submitted for review and possible publication on May 24, 2007; approved techniques. In these approaches one tries to study, model, and
on February 27, 2008. This paper is part of the Journal of Irrigation and analyze very complex phenomena, for which more-precise scien-
Drainage Engineering, Vol. 134, No. 6, December 1, 2008. ©ASCE, tific tools of the past were incapable of giving low-cost, analytic,
ISSN 0733-9437/2008/6-847–852/$25.00. and complete solutions. Neural networks, fuzzy systems, and evo-

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / NOVEMBER/DECEMBER 2008 / 847

J. Irrig. Drain Eng., 2008, 134(6): 847-852

 Fig. 1. Arbitrary cross section of an open channel
Downloaded from ascelibrary.org by University of Southern Queensland on 05/11/16. Copyright ASCE. For personal use only; all rights reserved.

lutionary computation are important areas of soft computing. An In this study, the minimum value of the objective function is
evolutionary algorithm 共EA兲, a subset of evolutionary computa- definite; limitations in time or fitness are not appropriate criteria
tion, is a generic population-based metaheuristic optimization al- for stopping the algorithm and the algorithm stops when the num-
gorithm. An EA uses mechanisms inspired by biological ber of generations reaches a predefined value.
evolution: reproduction, mutation, recombination, natural selec-
tion, and survival of the fittest. Candidate solutions for the opti-
Basic Equations
mization problem play the role of individuals in a population, and
the cost function determines the environment within which the Specific energy 共E兲 in an open channel is calculated from Eq. 共1兲
solutions “live.” Evolution of the population then takes place after
repeated applications of the above operators. v2 Q2
A genetic algorithm is one of the most popular classes of EA E = y cos2 ␪ + ␣ = y cos2 ␪ + ␣ 共1兲
2g 2gA2
used in soft computing. The common form of GA was described
by Goldberg 共1989兲. A GA is a stochastic search technique based
where ␣ = kinetic energy correction factor; y = flow depth;
on the mechanisms of natural selection and natural genetics. In
v = flow velocity; A = flow area; Q = discharge; and g
contrast to conventional search techniques, GAs start with an ini-
= gravitational acceleration 共Fig. 1兲. Since the minimum specific
tial set of random solutions, which are called parent populations
energy occurs at the critical point, the value of the critical depth
of the first generation. Each individual in the population is de-
can be found by taking the first derivative of Eq. 共1兲 with respect
noted as a chromosome, representing a solution to the problem. A
to y
chromosome is a string of symbols that is usually, but not neces-
sarily, a binary bit string. In recent years, GAs have been applied

冤 冥
to numerous civil engineering problems, particularly relating to dA dA
− 2A ␣Q2
system optimization or calibration 共Wang 1991; Balascio et al. dE ␣Q2 dy dy
= cos2 ␪ + = cos2 ␪ − 共2兲
1998; Karpouzos et al. 2001; Bozorg-Hadad and Sharifi 2005; dy 2g A4 gA3
Bakhtiari et al. 2007兲
The procedure of the genetic algorithm is outlined here: Using dA = Tdy cos ␪ and setting Eq. 共2兲 equal to zero to find the
1. The algorithm begins by creating a random initial population. minimum E, we have
2. The algorithm then creates a sequence of new populations 共or
children兲. At each step, the algorithm uses the individuals in ␣Q2T
the current generation to create the next generation. To create =1 共3兲
gA3 cos ␪
the new generation, the algorithm performs the following
steps:
where T = width of the flow surface. Eq. 共3兲 is the basic relation
a. Scores each member of the current population by com-
for calculation of critical depth in open-channel flow. If the slope
puting its fitness value.
of the channel is small 共less than 10%兲 and flow is uniform,
b. Scales the raw fitness scores to convert them into a
cos ␪ = 1 and ␣ = 1 are acceptable assumptions. Thus, Eq. 共3兲
more usable range of values.
becomes
c. Selects parents based on their fitness.
d. Produces children from the parents. Children are pro-
duced either by making random changes to a single Q 2T
=1 共4兲
parent—mutation—or by combining the vector entries gA3
of a pair of parents—crossover—or from individuals in
the current generation with the best fitness values— Knowing the discharge and channel geometry, the critical depth
elite children. can be obtained from Eq. 共4兲. For the majority of cross sections,
e. Replaces the current population with the children to where A and T are complex functions of y, the equation needs to
form the next generation. be solved numerically.
3. Generally, the algorithm stops when one of the following
stopping criteria is met:
Critical Depth Using GA
a. Number of generations.
b. Time limit in seconds. In order to find the critical depth using a GA, Eq. 共4兲 should be
c. Fitness limit. transformed into an objective function or

848 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / NOVEMBER/DECEMBER 2008

J. Irrig. Drain Eng., 2008, 134(6): 847-852

 F= 冏 Q2 ⫻ T
g ⫻ A3
−1 冏 共5兲

It is completely conspicuous that minimization of F will lead to

finding critical depth, hence

再 min F
constraint: 0 ⬍ y ⬍ y max
冎 共6兲

The above problem has just one constraint, depth, and the effect
of this constraint is incorporated into the GA using a penalty
function. If the generated depth is more than the maximum depth
Downloaded from ascelibrary.org by University of Southern Queensland on 05/11/16. Copyright ASCE. For personal use only; all rights reserved.

of channel, the penalty function will assign a penalty value to F.

Consequently, F will have a low score and the corresponding
value of y will not be used for the next generation and will van-
ish. To solve Eq. 共6兲, a model is developed using the following
input data:
1. The X and Y coordinates of several points on the channel
perimeter representing the channel geometry.
2. Constant value for g.
3. Discharge.
Then, the model calculates the section flow area 共A兲 and free
surface width 共T兲 to obtain a value of the objective function F.
Subsequently, the value of F is returned to the GA and a new
generation is created. This process continues until stopping crite-
ria terminate the algorithm.
In this study, the mentioned computations were performed in
the standard MATLAB environment 共MathWork 2004兲. Further-
more, because of the availability of GA routines in almost all
evolutionary references, there is no limitation about the computer
language and this method can be usable in any programming
language. A flowchart of the detailed process is shown in Fig. 2.

Numerical Results and Discussions

In this section, three channel shapes are used to demonstrate the

Fig. 2. Flowchart of methodology
capacity of our GA model: First, regular cross sections such as
triangular, rectangular, trapezoidal, and circular cross sections
were used. Second, more-complicated man-made sections such as
jaw and egg-shaped sections were used. Finally, a natural river higher value, but is still less than 2%. The sensitivity analysis
was chosen as a general case for a complicated section. In each shows that by increasing the number of generations the compari-
case, comparisons between the conventional solution and the pro- son index does not improve much. The fluctuations in the results
posed solution will be presented. are characteristic of the GA method.

Regular Cross Sections Egg and Jaw Sections

In Table 1, comparisons between the values of critical depth for Although they are nonprismatic, egg and jaw cross sections are
some regular cross sections from the presented model and ex- popular in sewerage and surface water networks when the dis-
plicit, graphical, or empirical relations are tabulated. The number charge is large. Egg shapes are commonly used in situations when
of generations and chromosomes for each shape are varied with the discharge varies a lot, while for jaw shapes the discharge is
respect to the complexity of the cross section. Differences be- generally more uniform and constrained for deep pipe. In Fig. 4,
tween the results, from a practical engineering point of view, are the jaw and egg cross sections used in calculation are shown.
negligible and the accuracy of the model is conspicuous. For an egg-shaped channel with a discharge of 0.4 m3 / s, the
Sensitivity analysis was carried out with respect to the number critical depth from the model after 30 generations with 50 chro-
of generations when the number of chromosomes was kept con- mosomes was calculated as 0.441 m; while using the graphical
stant 共number of chromosomes was chosen equal to 20兲. In this trial and error method gave 0.437 m. In order to check the sensi-
case, the relative difference between the model and the conven- tivity of the answer with respect to the number of generations, the
tional solution was chosen as a comparison index. In Fig. 3, the problem also was solved with 5, 10, 20, and 30 generations with
relative difference between the model and the conventional solu- 50 chromosomes. The results of the graphical trial and error so-
tion versus the number of generations is plotted for the cross lution, showed 17 trials before reaching the final value. In Fig. 5,
sections. For triangular and rectangular sections the index is close the fitness value 共amount of objective function兲 versus the number
to zero, and the worst case comparison index is less than 0.2%. In of iterations or number of generations is plotted.
trapezoidal and circular sections the comparison index has a In Fig. 6, corresponding values of depth in each iteration or

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / NOVEMBER/DECEMBER 2008 / 849

J. Irrig. Drain Eng., 2008, 134(6): 847-852

 Table 1. Critical Depth Using GA and Other Methods for Regular Cross Sections
GA method 共present兲
Critical depth by explicit
Cross Discharge or empirical method Critical depth Number of Number of
section Geometrical specifications 共m3 / s兲 共m兲 共m兲 generations chromosomes

z⫽1.5 3 0.96002* 0.9604 40 10

b⫽ 2 m 3 0.86047* 0.86043 60 10
Downloaded from ascelibrary.org by University of Southern Queensland on 05/11/16. Copyright ASCE. For personal use only; all rights reserved.

b⫽6 m
17 0.84** 0.84616 80 20
z⫽2

d0⫽2 m 5 1.06*** 1.07756 100 50

Note: *⫽explicit solution; **⫽graphical solution; ***⫽straub semiempirical solution.

generation are shown. Comparison between the two methods Natural Channel
shows superiority of the GA solution technique. The difference
As mentioned previously, the advantage of the GA method is that
between the trial and error solution and the GA solution is less
it can be generalized for any arbitrary cross section. To check the
than 1%, and this again shows the high performance of this
ability of this method, a general cross section of a river, as shown
method.
in Fig. 9, is chosen. Any general cross section can be illustrated as
Using the same specifications 共discharge= 0.4 m3 / s兲, the prob-
a series of straight lines with different slopes.
lem was solved for the illustrated jaw-shaped cross section of Fig.
For a flow of 5 m3 / s, setting the stopping criteria as 150 gen-
4. The critical depth from the model after 30 generations with 50
erations and the number of chromosomes as 50, critical depth will
chromosomes was equal to 0.228 m. A trial and error solution
be equal to 1.10636 m. In Fig. 10, the procedure of minimization
gave this value equal to 0.219. In Figs. 7 and 8, comparisons
for the objective function for this example is shown. Dot points
between the conventional and GA-based solutions for jaw shape
are best fitness in each generation and circle points are the aver-
section are shown.
age fitness value in each generation. High values of average fit-
ness in some generations are due to the penalty function.

Fig. 3. Relative percent of difference between conventional and GA

solution versus number of generations for regular cross sections
Fig. 5. Fitness value versus number of iterations or number of gen-
erations for egg shape

Fig. 6. Corresponding value of depth in each iteration or generations

Fig. 4. Jaw and egg cross sections for egg shape

850 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / NOVEMBER/DECEMBER 2008

J. Irrig. Drain Eng., 2008, 134(6): 847-852

 Fig. 7. Fitness value versus number of iterations or number of gen-
erations for jaw shape
Downloaded from ascelibrary.org by University of Southern Queensland on 05/11/16. Copyright ASCE. For personal use only; all rights reserved.

The same problem was solved by HEC-RAS 共USACE 2002兲.

This software, as one of the most widely used, uses two methods
for calculation of critical depth. The algorithm, which was used in
solving this problem, involved fitting a quadratic equation
through the specific energy associated with three depths. Using a
minimizing technique for this quadratic, the critical depth can be
found. The value of critical depth calculated by HEC-RAS was Fig. 10. Procedure of minimization of objective function by GA
1.08 m. In Fig. 11, flow specifications such as level of water, level during 150 generation 共river problem兲
of critical depth, and level of energy grid line are shown. The
difference between the GA and HEC-RAS solutions was less than
2%. The reasons for the differences are due to the approximations bolic method, equation fitting enters approximations in calcula-
made in the energy-depth equation, minimization of the quadratic tions and these approximations can reduce precision. The present
equation in HEC-RAS, or the absence of the kinetic energy cor- GA method has this ability to solve Eq. 共3兲 directly, which will
rection factor in the GA model. improve the accuracy of results. Furthermore, this method 共GA
Since Eq. 共3兲 is complicated to solve, using HEC-RAS some- method兲 is a more general approach for solving complicated
times cannot obtain convergent results. Basically, two approxima- equations, such as Eq. 共3兲, with complicated boundary conditions.
tions have been used in HEC-RAS to solve Eq. 共3兲 if the Thus, it is not limited to the calculation of critical depth. Concepts
convergence of models cannot be achieved, i.e., parabolic and of this method can be used in similar problems.
secant methods. In this regard, HEC-RAS fits a quadratic through
the specific energy associated with three depths. Using a minimiz-
ing technique for this quadratic, the critical depth may be found. Conclusions
However, it is an approximation, not exact solution. In the para-
In this study, the calculation of critical depth, an important pa-
rameter in hydraulic engineering, for several kinds of open-
channel cross sections, such as simple, regular, natural, and
tortuous shapes, based on one of the main branches of soft com-
puting, i.e., evolutionary computations, was presented. This
method is general, precise, and faster in comparison with the
previous methods.
The main advantage of the GA model is that it provides a
precise and faster prediction of critical depth than other previous
analytical or computational methods 共Straub 1978; Chadudhry
and Bhallamudi 1988兲. As mentioned previously, the commonly
used software, HEC-RAS, solves Eq. 共3兲 indirectly; i.e., it fits an
Fig. 8. Corresponding value of depth in each iteration or generations
equation to the graph of specific energy, and then by minimization
for jaw shape
of fitted equation, critical depth will be calculated. The fitting
procedure and minimization enter approximations into calcula-

Fig. 9. Arbitrary cross section of a river Fig. 11. Channel and flow specifications resulted by HEC-RAS

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / NOVEMBER/DECEMBER 2008 / 851

J. Irrig. Drain Eng., 2008, 134(6): 847-852

 tions and these approximations will reduce the precision of the depth in symmetrical compound channels.” J. Hydraul. Res., 26共4兲,
results. Unlike previous approaches, because of the ability of the 377–395.
GA approach in direct solving of Eq. 共3兲, it provides a more Dey, S. 共2001兲. “Flow measurement by the end-depth method in inverted
accurate prediction. Numerical examples demonstrated the capac- semicircular channels.” Flow Meas. Instrum., 12共4兲, 253–258.
ity of the present GA model. French, R. 共1986兲. Open-channel hydraulics, McGraw-Hill, New York.
It is noted that the concept of the GA approach present here Goldberg, D. 共1989兲. Genetic algorithms in search, optimization machine
also can be used for solving other hydraulic problems. This ap- learning, Addison-Wesley, Reading, Mass.
Karpouzos, D. K., Delay, F., Katsifarakis, K. L., and de Marsily, G.
proach may be generalized for solving other similar equations and
共2001兲. “A multipopulation genetic algorithm to solve the inverse
problems like normal depth.
problem in hydrogeology.” Water Resour. Res., 37共9兲, 2291.
Kurbanov, S. O., and Khanov, N. V. 共2004兲. “Computation of critical
depth of channels with polygonal profile.” J. Power Technol. Eng.,
Acknowledgments
Downloaded from ascelibrary.org by University of Southern Queensland on 05/11/16. Copyright ASCE. For personal use only; all rights reserved.

38共2兲, 71–73.
MathWorks. 共2004兲. Genetic algorithm and direct search toolbox user’s
The writers are grateful for the assistance from Mr. Farzad guide, The Mathworks Inc., Natick, Mass.
Faridafshin at Chalmers University of Technology. Smith, K. 共1972兲. “Computer determination of critical depth control
points in open channel flow.” Proc., Insitute of Civil Engineers, Part
2, Vol. 53, Institute of Civil Engineers, London, 461–470.
References Straub, W. O. 共1978兲. “A quick and easy way to calculate critical and
conjugate depths in circular open channels.” Civ. Eng. (N.Y.), 48共12兲,
Bakhtiari, M., Kanani, A., and Borghei, S. M. 共2007兲. “Normal depth in 70–71.
open channels based on genetic algorithm.” Proc., 32nd Int. Associa- Swamee, P. K., and Rathie, P. N. 共2005兲. “Exact equations for critical
tion of Hydraulic Engineering and Research, Venice. depth in a trapezoidal canal.” J. Irrig. Drain. Eng., 131共5兲, 474–476.
Balascio, C. C., Palmeri, D. J., and Gao, H. 共1998兲. “Use of a genetic U.S. Army Corps of Engineers 共USACE兲. 共2002兲. Hydraulic reference
algorithm and multiobjective programming for calibration of a hydro- manual, Institute For Water Resources, Hydrological Engineering
logical model.” Trans. ASAE, 41共3兲, 615–619. Center, USACE, Washington, D.C.
Bozorg-Hadad, O., and Sharifi, F. 共2005兲. “Genetic algorithm in optimal Wang, Q. J. 共1991兲. “The genetic algorithm and its application to cali-
design of stepped spillway and its down stream energy dissipaters.” brating conceptual rainfall-runoff models.” Water Resour. Res., 27共9兲,
Proc., 73rd Annual Meeting of the Int. Commission of Large Dams, 2467–2471.
Paper 058-OT. Wang, Z. 共1998兲. “Formula for calculating critical depth of trapezoidal
Chaudhry, M. H., and Bhallamudi, S. M. 共1988兲. “Computation of critical open channel.” J. Hydraul. Eng., 124共1兲, 90–91.

852 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / NOVEMBER/DECEMBER 2008

J. Irrig. Drain Eng., 2008, 134(6): 847-852