Anda di halaman 1dari 8

Flooding Probability Constrained Optimal Design

of Trapezoidal Channels
Amlan Das1

Abstract: Trapezoidal open channels are designed for specified slope, specific lining materials, and for a specified design flow. These
input design parameters are prone to uncertainties. The freeboard is used to confine the flow within the limits of the channel cross section
and to accommodate the uncertainty effects. Because of the size of uncertainty, the flow can overtop the freeboard with a “flooding
probability.” To design an optimum channel cross section for safety against overtopping, the flooding probability constrained optimum
channel design concept is introduced. An optimization model was developed which has two objective functions of minimizing the total
cost of the channel and minimizing the flooding probabilities subject to uniform flow equation as constraint. The constraint method of
multiobjective optimization is used in which the objective of minimizing the flooding probability is converted to a constraint, and the
resulting single objective optimization problem is solved. The flooding probability constraint is developed by using the first order analysis.
The final single objective optimization model is highly nonlinear and requires the use of the projected augmented Lagrangian technique.
The optimization model is applied for two cases, namely, channels having composite roughness and channels having uniform roughness.
The solution results of the analysis established the potential of the model.
DOI: 10.1061/共ASCE兲0733-9437共2007兲133:1共53兲
CE Database subject headings: Open channels; Roughness; Floods; Optimization; Lining.

Introduction the freeboard with a “flooding probability.” This paper documents


the development of the optimization model for channel design
Trapezoidal open channels are widely used as irrigation, drainage, that takes the flooding probability into account.
water transmission, and power canals. These channels are laid on To reduce cost and improve the aesthetic appearance, the chan-
a specified slope, fabricated by using specific lining materials, and nel is usually fabricated by using locally available cheaper con-
designed to convey a specified design flow. These input design struction materials along the side of the channel and more costly
parameters are prone to various uncertainties. The inaccurate con- cement concrete on the bed. Chow 共1959兲 defined these channels
trol at the off take point of the channel, and lateral inflow caused as composite channels in which roughness varies along the wetted
by variable surface and subsurface runoff, introduce uncertainty perimeter. The open channel cross-sectional design is generally
in the estimation of design flow. The actual roughness values of based on a one-dimensional analysis of steady flow in which the
the lining materials after fabrication may differ from the mean composite roughness is conventionally expressed as an equivalent
roughness values assumed in the design. The freeboards are used Manning “n.” Yen 共2002兲 listed 17 equivalent roughness formu-
to confine the flow within the limits of the channel cross section lations out of which the equivalent roughness expression of
and to accommodate the uncertainties. A channel design is a suc- Horton 共1933兲 and Einstein 共1934兲 present a case in which the
cess so long as the flow does not overtop the freeboard limits. A channels having uniform roughness become a special case of
good design seeks to maintain the economy and success of the channels with composite roughness. Das 共2000a兲 used the Horton
canal throughout its period of operation. The traditional optimal 共1933兲 and Einstein 共1934兲 equivalent roughness equation, and
design seeks to determine the optimum geometric dimensions of a Jain et al. 共2004兲 used Lotter’s 共1993兲 equation for optimal design
cross section that can carry the specified design flow for assumed of a trapezoidal cross section with composite roughness. Guo and
values of channel roughness along the channel of specified cross- Hughes 共1984兲 used freeboard as an input variable in the optimal
sectional shape and size, longitudinal bed slope, and other im- design of open channels. Monadjemi 共1994兲, Froehlich 共1994兲,
posed limitations on some geometric dimensions. In this process a and Das 共2000a兲 used the Lagrange multiplier method of con-
freeboard is needed to accommodate the uncertainties. However, strained optimization technique in optimal design of channels.
because of the size of the uncertainties, the flow can still overtop Das 共2000a兲 presented computational algorithms for solving the
system of simultaneous equations that result from the application
1
Assistant Professor, Dept. of Civil Engineering, National Institute of of the Lagrange multiplier method in optimal design of channels.
Technology, Durgapur 713209, West Bengal, India. These studies considered deterministic optimal design.
Note. Discussion open until July 1, 2007. Separate discussions must The Manning equation is widely used to estimate the uniform
be submitted for individual papers. To extend the closing date by one flow in an open channel 共Chow 1959兲. According to Manning’s
month, a written request must be filed with the ASCE Managing Editor.
equation, there exists one unique discharge for a given set of
The manuscript for this paper was submitted for review and possible
publication on August 2, 2005; approved on May 17, 2006. This paper is values of Manning’s n, longitudinal bed slope, side slope, bed
part of the Journal of Irrigation and Drainage Engineering, Vol. 133, width, and flow depth. Given that other parameters and geometric
No. 1, February 1, 2007. ©ASCE, ISSN 0733-9437/2007/1-53–60/ dimensions are specified, the flow depths increase when either or
$25.00. both the design flow and the Manning’s n increase. The flow and

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007 / 53

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
mean equivalent roughness; 共3兲 the Manning’s n values, longitu-
dinal slope values, design flow, and flow depths are probabilistic;
共4兲 the Manning’s n values, longitudinal slope values, and design
flow are mutually independent and flow depth is dependent on
these mutually independent variables; 共5兲 all randomly variable
parameters follow normal distribution; and 共6兲 the first order un-
certainty analysis method given in Chow et al. 共1988兲 depicts the
random variations. The objective function seeks to minimize the
total construction cost per unit running length of channel cross
section that is obtained by summing up the cost for the cross-
sectional area, cost for lining the two side faces using two differ-
ent lining materials, and cost for lining the bed. These costs apply
Fig. 1. Definition sketch for trapezoidal cross section having for the total cross section that includes the wetted part and free-
composite roughness board part. The overall objective function is expressed as
F共x兲 = c1A f + c2 P1f + c3 P2f + c4 P3f 共1兲
Manning’s n are random variables and the randomness arises in which A f ⫽total area of the cross section; c1⫽per unit area cost
from uncertainties associated with their values. In the literature, for cross-sectional area; c2⫽per unit length cost for perimeter P1f ;
the probabilistic design of open channels are addressed by Easa c3⫽per unit length cost for perimeter P2f ; c4⫽per unit length cost
共1992兲. The risk analysis for hydraulic design is addressed by for the perimeter P3f ; F共x兲⫽objective function expressed as a
Tung and Mays 共1980兲, Lee and Mays 共1986兲, and Chow et al. function of x; and x⫽vector dependent on b, y, Z1, and Z2. The
共1988兲. The deterministic optimum channels are also prone to first constraint of the model is the Manning’s uniform flow equa-
variation in design parameters. A flooding probability analysis is tion describing the mean channel capacity which may be ex-
performed before formulating a flooding probability constrained pressed as 共Das 2000a兲
optimum design model, and a part of the present study is devoted
A5/3
to this end. Q = 冑S0 共2兲
The traditional risk analysis is used to determine the probabil-
ity of overtopping the freeboard limits when the input parameters
共兺 n 3/2
i Pi 兲
2/3

equal or exceed a specified upper limit of a given probability. The where Q⫽design flow; S0⫽longitudinal bed slope;
chance constrained design seeks to obtain channel dimensions A⫽cross-sectional area of flow; Pi⫽flow wetted perimeters
that have extra allowances for the variation of specified probabili- 共i = 1 , 2 , 3兲; and ni⫽Manning’s n values 共i = 1 , 2 , 3兲. The next con-
ties in design flow, channel roughness, and other parameters over straint is the chance constraint. In real life situation, the actual
the mean design values. The flooding probability constrained op- flow may exceed the design flow because of probable fault in the
timal design seeks to obtain the optimum dimensions of the chan- control of flow at the off take point, and the uncertain lateral
nel cross section along with the specified freeboard in a manner inflow. The actual Manning roughness values may exceed the
that the flow cross section will just convey the design flow and assumed design values because of imperfections in fabrication.
the freeboard will just accommodate with a specified flooding The physical bed slope achieved may differ from the design val-
probability. The present study introduces this concept of flooding ues because of fabrication faults. These variations can cause the
probability constrained optimal design of open channels. occasional flooding of man made open channels. To maintain pro-
The objective of the present study is to develop a model for visions for these possible variations within the designed cross-
flooding probability constrained optimal design of channels hav- sectional dimensions, a flooding probability constraint must be
ing composite roughness that addresses the random variation of incorporated in the optimal design model. In this study, the flood-
design input parameters. To ensure that the channel flow will not ing probability constraint is developed by using the first order
overtop the freeboard with a certain probability, a flooding prob- analysis that essentially uses the calculus based differentiation of
ability constraint is added to the optimization model. The flooding the Manning uniform flow Eq. 共2兲 with respect to the random
probability constraint is developed by considering the random variables. To include the effect of change in design flow, the
variation of the composite Manning’s n values along the channel following derivative of Eq. 共2兲 is required.

冋 册
perimeter for the design flow, flow depths, and longitudinal
slopes. The incorporation of a flooding probability constraint
dQ
=Q
5 1 dA 2

1 d
共兺 n 3/2
i Pi 兲 共3兲
within the traditional optimization models for channel cross- dy 3 A dy 3 共兺 n3/2
i Pi 兲 dy
sectional design is the main contribution of the present study.
To account for the uncertain chance of variation of bed slopes, the
following derivative is needed:
Mathematical Formulation dQ Q
= 共4兲
Fig. 1 shows a definition sketch of the trapezoidal cross section dS0 2S0
having composite roughness. In this definition sketch: b⫽bed The following derivative is to account for the effects of variation
width; y⫽flow depth; Z1 and Z2⫽side slopes; n1, n2, and of Manning’s roughness:

冋 共兺 册
n3⫽ Manning’s n values for perimeters P1f , P2f , and P3f , respec-
tively; and f⫽freeboard above the free water surface. The present dQ
=−Q
2 1 d
共兺 n 3/2
i Pi 兲 共5兲
optimization model is based on the assumptions that: 共1兲 the cost
parameters are deterministic; 共2兲 the Horton 共1933兲 and Einstein
dni 3 n3/2
i Pi 兲 dni

共1934兲 equivalent roughness formula is valid for computation of The following results from application of basic calculus:

54 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
dy
=
dQ dQ
1
共6兲 S2y = 冉 冊 冉 冊 冉 冊 冉 冊
dy
dQ
2
SQ2 +
dy
dn1
2
S2n1 +
dy
dn2
2
S2n2 +
dy
dn3
2
S2n3

dy
+冉 冊 dy

冉 冊

dS0
冉 冊 冉 冊2
S2S0 =
1
dQ 2 SQ2 +
dQ
dn1
2
S2n1 +
dQ
dn2
2
S2n2

dy dy dQ 1 dQ dy

冉 冊 冉 冊 册
= = 共7兲
dni dQ dni dQ dni dQ 2
dQ 2

dy + S2n3 + S2S0 共9兲


dn3 dS0
which may be rearranged to give the following:

冋 冉 冊 冉 冊 冉 冊
dy dy dQ 1 dQ
= = 共8兲 dQ 1 2 dQ 2
dQ 2
dQ 2
dS0 dQ dS0 dQ dS0 = SQ + S2n1 + S2n2 + S2n3
dy Sy dn1 dn2 dn3

冉 冊 册
dy
2 1/2
dQ
According to the fourth assumption, the variance of y 共=Sy兲, is + S2S0 共10兲
dS0
dependent on the variance of mutually independent variables
Q 共=SQ兲, n1 共=Sn1兲, n2 共=Sn2兲, n3 共=Sn3兲, and S0 共=SS0兲 and is ex- After substituting dQ / dy from Eq. 共3兲 in Eq. 共10兲 the following is
pressed as 共Chow et al. 1988兲 written:

5 dA

A dy 共

2 d
n P 兲 dy
3/2
共兺 n 3/2
i Pi 兲 = S3 再再 冋 共兺
SQ2 4
+
Q2 9
1 d
ni Pi dn1
3/2

共兺 n 3/2
i Pi 兲 册 冊 冋 共兺
2
S2n1 +
4
9
1 d
ni Pi dn2
3/2

共兺 n 3/2
i Pi 兲 册 2
S2n2


i i y

+
4
9 冋 共兺 1
n3/2 兲
d
dn3
共兺 n 3/2
i Pi 兲 册 2
S2n3 +
1
4S20
S2S0
1/2

再再
i Pi

=
3
Sy
SQ2
Q2 9
+
共兺 n
4
3/2 2

再冋 d 3/2
dn1
2
共n1 P1兲 S2n1 +
d 3/2
dn2
册 册册 冋
2
共n2 P2兲 S2n2 册

i Pi

+ 冋 d 3/2
dn3
2
1

共n3 P3兲 S2n3 + 2 S2S0
4S0
1/2

=
3
Sy 再再 SQ2
Q2
+
共兺 n
1
3/2
i Pi 兲 册
2 关n1 P1Sn1兴 +关n2 P2Sn2 + n3 P3Sn3兴 +
2 2 2 2 2 2
4S0
1 2
2 S S0 冎 1/2
共11兲

Finally, the following compact form of the flooding probability Two specific cases of Eq. 共12兲 are of interest to practical en-
constraint is written: gineering. Case 1 where n1 ⫽ n2 ⫽ n3 represents the ideal case, in
which the chance constraint takes the form as given in Eq. 共12兲.
Case 2 where n1 = n2 = n3 represents channels with all faces having
5 dA

2 d
共兺 n 3/2
i Pi 兲
共兺 兲
identical roughness. For this case terms containing n2 and n3 get
A dy n3/2 dy

再 冎
i Pi dropped, P1 and P1f capture the complete perimeter information,
3 SQ2 1 1 1/2 c2 = c3 = c4, and Eq. 共2兲 is written as
= + 兺 ni P2i Sn2i + 2
2 S S0 共12兲
Sy Q2 共兺 n 3/2
i Pi
2
兲 4S0
冑S0 A5/3
The constraint 共12兲 represents the deterministic equivalent of the Q= 共14兲
n1 P2/3
1
actual probabilistic constraint that is represented as
and Eq. 共12兲 is written as
P共y ⬎ y + f兲 = K1 共13兲
which literally means that the probability of exceedence of flow
depth over the freeboard is a fixed value K1. Because of the con-
straint 共13兲 the overall optimization problem is a probabilistic
5 dA 2 d

A dy P1 dy
共P1兲 = +
2

+
2
3 SQ2 Sn1 SS0
Sy Q2 n21 4S20
再 冎 1/2

共15兲

optimization problem. To obtain a solution, the deterministic ver- For the general case of composite roughness, Eqs. 共2兲 and 共12兲
sion of the problem is solved, for which a procedure for obtaining represent the constraints of the optimization model. The complete
the values of variances must be explained and is given optimization problem is expressed as
separately. Minimize

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007 / 55

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
F共x兲 = c1A f + c2 P1f + c3 P2f + c4 P3f 共16兲 C2共x兲 at the current iterate xi. ␳⫽scalar penalty parameter. The
subproblems are solved by using the reduced-gradient algorithm
Subject to in conjunction with a quasi-Newton algorithm. MINOS 共Murtagh
1. Uniform flow constraint and Saunders 1993兲, a standard implementation of the aforesaid
Q A5/3 computational algorithm that was used in Das 共2000b兲, was used
C1共x兲 = − =0 共17兲
冑S0 共兺 n 3/2
i Pi
2/3

to solve the present optimization problem.

2. Deterministic version of flooding probability constraint


Procedure of Obtaining Values of Variances
C2共x兲 =
5 dA

2 d
共兺 n3/2
i Pi 兲
A dy 共兺 ni Pi dy
3/2
兲 Chow et al. 共1988兲 defined the standard normal variable Z as


3
Sy 再 SQ2
Q2
+
共兺
1
n3/2
i Pi
2

兺 ni P2i Sn2i +
1
4S0
2
2 S S0 冎 1/2
=0 Z=
X−␮

where X⫽value of the random variable; ␮⫽mean value of X;
共22兲

共18兲
␴2⫽variance of X. In sample statistic, ␮ = X̄, and ␴ = SX. For a
Here, C1共x兲 and C2共x兲 represent the two constraints as functions given value of Z, the standard normal probability distribution
of x 共i.e., b , y , Z1, and Z2兲. The objective function 共16兲 together function value FZ共Z兲 is obtained from a table of cumulative prob-
with equality type constraints 共17兲 and 共18兲 constitute the deter- ability of the standard normal distribution 共see Chow et al. 1988兲.
ministic version of the actual probabilistic optimization problem. For Z ⬍ 0, one has to use FZ共Z兲 = 1 − FZ共兩Z 兩 兲 where
The non-negativity restrictions are also applied. Note that, drop- FZ共兩Z 兩 兲⫽tabulated value for 兩Z兩 and 兩Z兩⫽absolute value of Z. For
ping out of constraint 共18兲 will convert the optimization problem the present optimization problem, independent values of K1 for
to a deterministic version reported in Das 共2000a兲. Here, the ap- Q , y , n1 , n2 , n3, and S0 are specified which represent the 关1
plication of Lagrange multiplier 共LM兲 technique based methodol- − FZ共Z兲兴 values for the respective random variables. The FZ共Z兲
ogy of Monadjemi 共1994兲, Froehlich 共1994兲, and Das 共2000a兲 will and corresponding Z values are stored in array form. From speci-
give the required governing equations for the desired optimal so- fied values of K1, the Z values are determined by using linear
lutions and was attempted. However, the LM technique was interpolation of array of 关FZ共Z兲 , Z兴. The next step is to obtain the
found to be inadequate to solve this optimization problem be- values of the variances. As an example, for variable y , 兵共y + f兲
cause, by starting from any arbitrary starting point, the arrival to − y其 represents 共X − ␮兲. Therefore, Sy is obtained by using Eq.
a final optimal solution could not be ensured. A series of explor- 共22兲. The procedure is repeated for other random variables.
atory attempts were made to finally identify that the constraint
共18兲 makes the optimization problem highly nonlinear and the
LM technique based methodology performs unsatisfactorily for Flooding Probability Analysis of Deterministic
the present problem. Therefore, the projected augmented La- Optimum Channel Cross Section
grangian 共PAL兲 technique 共Murtagh and Saunders 1993兲 was
adopted. Starting from an initial guess solution, the PAL tech- The objective of this flooding probability analysis is twofold.
nique performs a sequence of linearly constrained subproblem First, it will reflect the vulnerability of the deterministic optimum
solving major iterations. In each major iteration the following channel cross section against the random variation of design input
optimization problem is solved: parameters and establish the need for a chance constrained de-
Minimize

冋冋 再 冎册
sign. Second, the present flooding probability constrained optimal

F共x兲 + ␭1i C1共x兲 − C1共xi兲 +


⳵ C1共x兲
⳵x
冏 共x − xi兲
design model can be verified for its correctness by using the re-
sults of flooding probability analysis. The flooding probability
analysis is performed for two specific cases of 共1兲 channels hav-

冋冋 再 冎册
x=xi

+ ␭2i C2共x兲 − C2共xi兲 +


⳵ C2共x兲
⳵x
冏 共x − xi兲
ing composite roughness, and 共2兲 channels having uniform rough-
ness. To perform this probability analysis, optimal channel cross

冋冋 再 冎册
x=xi sections are designed as a first step, thereafter, the actual flooding
1
+ ␳
2
C1共x兲 − C1共xi兲 +
⳵ C1共x兲
⳵x
冏 共x − xi兲
2 probability analysis is performed for the optimal cross section.

冋冋 再 冎册
x=xi


Flooding Probability Analysis of Optimum Channel
1 ⳵ C2共x兲 2
Cross Section Having Composite Roughness
+ ␳ C2共x兲 − C2共xi兲 + 共x − xi兲 共19兲
2 ⳵x x=xi An optimal open channel is to be designed for conveying a design
Subject to flow of Q = 100 m3 / s. The values of various costs are c1 = 0.50,

冏 冏
c2 = 0.3, c3 = 0.35, c4 = 0.4; mean values of channel roughness
⳵ C1共x兲 along the wetted perimeter are n1 = 0.030, n2 = 0.025, n3 = 0.020,
C1共xi兲 + 共x − xi兲 = 0 共20兲
⳵x x=xi
the freeboard to be provided is f = 0.5 m, the longitudinal bed
slope is S0 = 0.0025. The optimal design is performed by solving

C2共xi兲 + 冏 ⳵ C2共x兲
⳵x
冏 x=xi
共x − xi兲 = 0 共21兲
the following optimization model 共Das 2000a兲:
Minimize

F = c1A f + c2 P1f + c3 P2f + c4 P3f 共23兲


Here, ␭1i and ␭2i⫽estimate of the Lagrangian multipliers ␭1 and
␭2, respectively, that are associated with constraints C1共x兲 and Subject to uniform flow constraint

56 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
Q A5/3 case has a flooding probability of 38.27% which is also consid-
− =0 共24兲
冑S0 共兺 n 3/2
i Pi 兲 2/3 ered to be very high. In field situations, the channels should be
designed with lesser flooding probabilities. In this context, the
The final optimal solution gives the total cost per unit running methodology developed in the present study should be used.
length⫽22.935 currency units, and the channel dimensions as
y = 4.033 m, b = 6.404 m, Z1 = 0.253共H兲 : 1共V兲, and Z2
= 0.292共H兲 : 1共V兲. The flooding probability analysis is performed Flooding Probability Constrained Optimal Design
for random change in the parameter values to n1 = 0.033, of Channel Cross Section
n2 = 0.028, n3 = 0.023, S0 = 0.0028, and Q = 120 m3 / s, with 30, 25,
20, 30, and 35% exceedence probabilities, respectively. The The flooding probability analysis of deterministic optimum chan-
FZ共Z兲, Z, and SX values for these specified values of exceedence nels established the need for flooding probability constrained op-
probabilities are timal design of channel cross section in order to reduce the flood-
ing probability. A channel cross section designer may think to
FZ共Z兲n1 = 1 − P共n1 ⬎ 0.033兲 = 1 − 0.3 = 0.7; simultaneously minimize both the cost of the channel and the
flooding probability. Therefore, the present optimal design is vi-
Zn1 = 0.5244; Sn1 = 0.00572
sualized as a two objective optimization problem. The first objec-
tive is to design a minimum cost channel cross section. The sec-
FZ共Z兲n2 = 1 − P共n2 ⬎ 0.028兲 = 1 − 0.25 = 0.75; ond objective is to minimize the flooding probability. The
Zn2 = 0.6745; Sn2 = 0.00445 constraint is the uniform flow equation. This two objective opti-
mization model is formally expressed as
Minimize
FZ共Z兲n3 = 1 − P共n3 ⬎ 0.023兲 = 1 − 0.2 = 0.8;
Zn3 = 0.8418; Sn3 = 0.00356 J1 = c1A f + c2 P1f + c3 P2f + c4 P3f 共25兲
Minimize
FZ共Z兲S0 = 1 − P共S0 ⬎ 0.0028兲 = 1 − 0.3 = 0.7;
ZS0 = 0.5244; SS0 = 0.00057 J2 = P共y ⬎ y + f兲 共26兲
Subject to uniform flow constraint
FZ共Z兲Q = 1 − P共Q ⬎ 120兲 = 1 − 0.35 = 0.65;
ZQ = 0.3854; SQ = 51.89414 Q A5/3
C1 = − =0 共27兲
By inserting these values in Eq. 共12兲 and after solving the follow-
冑S0 共兺 n 3/2
i Pi 兲
2/3

ing are obtained:


Here, an element of probability is associated with the second
Sy = 1.4984; Zy = 0.3337; FZ共Z兲y = 0.6307; objective only, and hence it falls in the stochastic optimization
category. Also, this is a two objective vector optimization prob-
P共y ⬎ y + f兲 = 1 − FZ共Z兲y = 1 − 0.6307 = 0.3693
lem, and its solutions are expressed in the form of Pareto optimal
Thus, the deterministically designed optimal cross section for the solutions, i.e., the noninferior set of solutions. The constrained
present case of composite roughness has a flooding probability of method of multiple objective programming technique is used to
36.93% which is considered to be very high. To obtain an optimal obtain the Pareto optimal solutions for the two objective optimi-
cross section that has lesser flooding probabilities, the flooding zation problem of the present study. The constrained method
probability constrained optimal design model developed in the transforms the vector optimization problem to a scalar one. In the
present study could be used. present study, the transformation from vector to scalar optimiza-
tion is done by treating the first objective that seeks to minimize
the total cost of the channel as the single objective in the scalar
Flooding Probability Analysis of Optimum Channel
problem and the second objective that seeks to minimize the
Cross Section Having Uniform Roughness
flooding probability as an additional constraint in the form of
For this case of channels having uniform roughness 共n1 = n2 = n3兲 objective bound. The Pareto optimal solutions are obtained by
the deterministic optimization model is formulated by taking Eq. solving the scalar problem such that the new constraints 共i.e., the
共23兲 as an objective function and Eq. 共14兲 as a constraint along transformed objectives兲 are satisfied to desired levels 共i.e.,
with the non-negativity restraints. The specified data that are used bounds兲 in a progressive manner. In other words, parametric
in this deterministic optimal design are Q = 100 m3 / s, c1 = 0.50, variations of the objective bounds trace out the noninferior set of
c2 = c3 = c4 = 0.3, n1 = n2 = n3 = 0.030, f = 0.5 m, and S0 = 0.0025. The solutions 共Cohon and Marks 1975兲. The optimization problem
final results of the optimal design are total cost per unit running formulation given by the single objective function 共16兲, and two
length⫽24.963 currency units, y = 4.883 m, b = 5.406 m, Z1 constraints 共17兲 and 共18兲 represent the scalar-cum-deterministic
= 0.369共H兲 : 1共V兲, and Z2 = 0.369共H兲 : 1共V兲. The flooding probabil- equivalent of the vector-cum-stochastic optimization problem of
ity analysis is performed by using Eq. 共15兲 for random change in the present study, and is solved in parametric order.
the parameter values to n1 = n2 = n3 = 0.033, S0 = 0.0028, The two objectives J1 and J2 may conflict each other. To
Q = 120 m3 / s, with 30, 30, and 35% exceedence probabilities, re- explore the possible existence of any conflict between the objec-
spectively. By performing the flooding probability analysis in a tives J1 and J2, the developed optimization model is applied to
manner identical to that of the composite roughness case, two examples, namely, 共1兲 the case of channels having composite
P共y ⬎ y + f兲 = 0.3827 is obtained. Thus, the deterministically de- roughness, and 共2兲 the case of channels having uniform
signed optimal cross section for the present uniform roughness roughness.

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007 / 57

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
duced to 0.35 the total cost increases to 23.066 and it opposes the
objective of minimizing the total cost. The present results show
that there is a trade off between these two objectives for flooding
probabilities less than 0.3693. The flooding probability of 0.3693
is resulted from deterministic optimum channel design in the
flooding probability analysis presented earlier. At this flooding
probability the total cost of the channel is minimized and repre-
sents an extreme point of the trade off relationship between the
two objectives. Intuitively, this verifies the correctness of the de-
veloped optimization model. In order to use the results of these
Pareto optimal solutions, the user’s utility function, i.e., a math-
ematical function formulated by incorporating the money valued
weightage of the individual objective function, must be made
available by the policy makers.
The general trend of the results in Fig. 2 are that the bed width
required for an optimum cross section increases as the flooding
probability reduces, and the required flow depth decreases as the
flooding probability also reduces. The specified freeboard value is
kept fixed at 0.5 m for the complete range of study. The above
particular trend of the result is attributed to the mathematical
expression of the flooding probability constraint. Also, intuitively,
the flooding probability of a channel will be less when its flow
depths are less and bed widths are large. Another trend is the
Fig. 2. Flooding probability constrained optimal solutions for required side slopes for an optimum channel increase with re-
composite roughness case duced flooding probability to a certain limit and thereafter it re-
duces. This particular trend may be attributed to the mathematical
character of the constraints and the objective functions. However,
the exact value of flooding probability at which the required side
Flooding Probability Constrained Optimal Design of slope values become maximum is not investigated because the
Channel Cross Section Having Composite Roughness particular issue may have academic importance only.
An optimal open channel is to be designed for conveying a design
flow of Q = 100 m3 / s. The mean values of various input param- Flooding Probability Constrained Optimal Design of
eters are c1 = 0.50, c2 = 0.3, c3 = 0.35, c4 = 0.4, n1 = 0.030, Channel Cross Section Having Uniform Roughness
n2 = 0.025, n3 = 0.020, f = 0.5 m, and S0 = 0.0025. Some of the input
parameter values randomly change to n1 = 0.033, n2 = 0.028, Here, an optimal open channel is to be designed for conveying a
n3 = 0.023, S0 = 0.0028, and Q = 120 m3 / s, with 30, 25, 20, 30, and design flow of Q = 100 m3 / s. The mean values of various input
35% exceedence probabilities, respectively. Explore the optimal parameters are c1 = 0.50, c2 = c3 = c4 = 0.3, n1 = n2 = n3 = 0.030,
design of the channel cross sections that minimize both the total f = 0.5m, and S0 = 0.0025. Some of the input parameter values
costs and the flooding probabilities. randomly change to n1 = n2 = n3 = 0.033, S0 = 0.0028, and
For the given input information, Sn1 = 0.00572; Sn2 = 0.00445; Q = 120 m3 / s, with 30, 30 and 35% exceedence probabilities, re-
Sn3 = 0.00356; SS0 = 0.00057; and SQ = 51.89414 are taken from spectively. The optimal design of the channel cross sections that
previous calculations and are used as additional input data for the minimize both the total costs and the flooding probabilities re-
final optimization model. To obtain the set of noninferior solu- quires study.
tions, a value of P共y ⬎ y + f兲 is assumed and corresponding Sy This explorative study is performed by following identical
value is calculated which constitute the final input parameter of procedures to that of the previous case. Here, the appropriate
the optimization model. P共y ⬎ y + f兲 is a continuous parameter, expressions for the uniform roughness case are used in the opti-
but, the noninferior set of solutions are obtained for its discrete mization formulation. The outcome of the study is shown in Fig.
values in the range of 0.025 to 0.475 in steps of 0.025. For each 3 in terms of the plot of Pareto optimal solutions and variation of
value of P共y ⬎ y + f兲 the scalar deterministic version of the biob- required geometric dimensions for given flooding probabilities for
jective optimization problem which is given by Eqs. 共16兲 through uniform roughness case in a single figure. The general trends are
共18兲 is solved to obtain one particular point in the Pareto optimal again observed to remain identical to that of the composite rough-
solutions. ness case except the specific numerical values.
Fig. 2 shows the Pareto optimal solutions for the present opti- The results again show the absence of any conflict between
mization problem. In Fig. 2 the variations of total cost, y, b, Z1, minimization of flooding probabilities and minimization of total
and Z2 are plotted against flooding probability in a single figure. cost for flooding probabilities greater than 0.3827, and the pres-
To minimize the total cost and the flooding probability, one has to ence of conflict between the two objectives for flooding probabili-
move toward the origin along the vertical and horizontal axes, ties less than 0.3827. The flooding probability of 0.3827 that
respectively, following the total cost variation curve. For flooding resulted from deterministic optimum channel design in the flood-
probabilities greater than 0.3693, there is no conflict between the ing probability analysis presented earlier represents one of the
two objectives. However, for flooding probabilities less than extremities of the zone of conflict between the two objectives.
0.3693, the two objectives conflict with each other. For example, The observations and conclusions about the results remain iden-
the total cost of the channel at flooding probability 0.3693 is tical to that of the composite roughness case except the numerical
22.935. In order to minimize the flooding probability if it is re- values.

58 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
and Mays 1980兲. Further, the identification of the best probability
distribution for flooding probability constrained optimal design of
open channels remains unexplored in this study. The use of other
relationships for equivalent Manning roughness expression can
also be a topic for future research. To solve the optimization
model a standard nonlinear optimization solver routine is used
because of the numerical difficulties with other existing methods,
and the presently used optimization routine is easily portable,
used to solve a large variety of nonlinear optimization problems,
robust and stable. The main objective of the present study is to
establish the need for the optimization model and the results of
the present study establish this. In this regard, scope exists to
develop simpler numerical methodologies for solving the devel-
oped optimization model.

Conclusions

Optimum design of open channels are effective in achieving the


overall economy open channel construction. All open channels
are subject to a flooding risk. The traditional deterministic opti-
mum channels are no exception. To design an optimum channel
Fig. 3. Flooding probability constrained optimal solutions for cross section for safety against flooding, the flooding probability
uniform roughness case constrained optimum channel design concept is introduced. The
optimal design model is developed as a two-objective nonlinear
optimization model that considers minimizing the total cost of the
The two aforesaid applications of the developed optimization channel and minimizing the flooding probabilities subject to
models are demonstrated for specified sets of input values and uniform flow equation as a constraint. The constraint method of
show the potential of the developed model for use in the design of multiobjective optimization is used in which the objective of
drainage channels. Extension of the study for other sets of input minimizing the flooding probability is converted to a constraint,
data requires routine application of the developed model and can and the resulting single objective optimization problem is solved.
be done easily. A fixed value of freeboard is used in the design as The effect of parametric variation of this converted constraint on
per common practice. The freeboard can be specified as a func- the channel design is investigated in this study. For the converted
tion of either flow depth 共Chow 1959兲 or flow velocity 共ASCE single objective optimization model, the flooding probability con-
1993兲. The incorporation of these types of freeboard specification straint is developed by using the first order analysis. The final
requires the substitution of the freeboard term by appropriate ex- single objective optimization model is highly nonlinear and re-
pressions in the formulation of the optimization problem. Further, quires the use of projected augmented Lagrangian technique. The
one may think to specify a probability and then determine the developed optimization model is applied for two cases, namely,
freeboard. In this regard, the Guo and Hughes 共1984兲 determinis- channels having composite roughness and channels having uni-
tic optimization results show that the freeboard must be specified form roughness. The following major conclusions are drawn from
as an input parameter in the optimization model. But the issue this study.
remains open for future studies.
• The two objectives of minimizing the flooding probability and
Simplified models for assessing the random variation of pa-
minimizing the total costs conflict with each other;
rameters are used in the present study. For field situations, the
• The deterministic optimum channel design is an extremity in
experience from a large number of existing channels need to be
this conflict;
collected and analyzed to obtain a general estimate of the random
• For lesser flooding probabilities, the channels of smaller flow
variability limits and the probability distributions. These
depths and larger bed width and steeper bank slopes are re-
estimates give a general guideline for assuming the probability
distribution. For example, the present study is based on the as- quired rather than the deterministic optimum channel;
sumption of validity of standard normal distribution that is used • The total costs increase as the flooding probability is reduced;
for many hydrological and hydraulic systems and is conservative • User’s utility function must be defined by the policy makers to
共Easa 1992兲. For practical application of the developed optimiza- use the present study results; and
tion model, the user specifies the probability values for which the • The developed optimization model has a potential for future
design is to be made. The value of the normalized variate corre- use.
sponding to the probability value is then obtained by following
the standard probability distributions for the assumed probability.
The limiting values of variation of design parameters are specified
either from previous records or user’s choice. The variance values Notation
are then obtained from Eq. 共22兲 which are used in the optimiza-
tion model. In this regard, scope exists to develop the methodol- The following symbols are used in this paper:
ogy by considering the use of other probability distributions such A ⫽ area of flow cross section;
as Poisson, Type I extremal, and lognormal distributions 共Tung A f ⫽ total area of cross section;

JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007 / 59

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
b ⫽ bed width; References
C1共x兲, C2共x兲 ⫽ two constraints as functions of x 共i.e., b, y,
Z1, and Z2兲; ASCE. 共1993兲. “Design and construction of urban stormwater manage-
c1 ⫽ per unit area cost for cross-sectional area; ment systems.” ASCE Manuals and Reports on Engineering Practice
c2 ⫽ per unit length cost for perimeter P1f ; No. 77, ASCE, ISBN 0872628558.
c3 ⫽ per unit length cost for perimeter P2f ; Chow, V. T. 共1959兲. Open channel hydraulics, McGraw-Hill, Singapore.
c4 ⫽ per unit length cost for perimeter P3f ; Chow, V. T., Maidment, D. R., and Mays, L. W. 共1988兲. Applied hydrol-
F共x兲 ⫽ objective function expressed as function of x; ogy, McGraw-Hill, Singapore.
FZ共Z兲 ⫽ standard normal probability distribution Cohon, J. L., and Marks, D. H. 共1975兲. “A review and evaluation of
function value; multiobjective programming techniques.” Water Resour. Res., 11共2兲,
f ⫽ freeboard above the free water surface; 208–220.
J1, J2 ⫽ two objectives; Das, A. 共2000a兲. “Optimal channel cross section with composite rough-
K1 ⫽ probability of exceedence of flow depth over ness.” J. Irrig. Drain. Eng., 126共1兲, 68–72.
the freeboard; Das, A. 共2000b兲. “Optimization based simulation and design of tile drain-
n1, n2, n3 ⫽ Manning’s n values; age systems.” J. Irrig. Drain. Eng., 126共6兲, 381–388.
Easa, S. M. 共1992兲. “Probabilistic design of open drainage channels.” J.
P1, P2, P3 ⫽ three perimeters of flow cross section;
Irrig. Drain. Eng., 118共6兲, 868–881.
P1f ,P2f ,P3f ⫽ three perimeters of full cross section;
Einstein, H. A. 共1934兲. “Der hydraulische oder profile-radius 关The hy-
P共. . . . . . . 兲 ⫽ probability of 共. . . . . . . . 兲;
draulic or cross-section radius兴.” Schweizerische Bauzeitung, Zurich,
Q ⫽ design flow;
103共8兲, 89–91 共in German兲.
Sn1 ⫽ variance of n1;
Froehlich, D. C. 共1994兲. “Width and depth-constrained best trapezoidal
Sn2 ⫽ variance of n2;
section.” J. Irrig. Drain. Eng., 120共4兲, 828–835.
Sn3 ⫽ variance of n3;
Guo, C. Y., and Hughes, W. C. 共1984兲. “Optimal channel cross section
SQ ⫽ variance of Q; with freeboard.” J. Irrig. Drain. Eng., 110共3兲, 304–314.
SS0 ⫽ variance of S0; Horton, R. E. 共1933兲. “Separate roughness coefficients for channel bot-
Sy ⫽ variance of y; tom and sides.” Eng. News-Rec., 111共22兲, 652–653.
S0 ⫽ longitudinal bed slope; Jain, A., Bhattacharjya, R. K., and Sanaga, S. 共2004兲. “Optimal design of
X ⫽ value of random variable; composite channels using genetic algorithm.” J. Irrig. Drain. Eng.,
X̄ ⫽ mean value of X in sample statistic; 130共4兲, 286–295.
x ⫽ vector made of b, y, Z1, and Z2; Lee, H. L., and Mays, L. W. 共1986兲, “Hydraulic uncertainties in flood
xi ⫽ current iterate; levee capacity.” J. Hydraul. Eng., 112共10兲, 928–934.
y ⫽ flow depth; Lotter, G. K. 共1933兲. “Soobrazheniia k Gidravlicheskomu Raschetu Rusel
Z ⫽ standard normal variable; s Razlichnoi Sherokho-vatostiiu Stenok 共Considerations on hydraulic
兩Z兩 ⫽ absolute value of Z; design of channels with different roughness of walls兲.” Izvestiia Vse-
Z1共H兲:1共V兲 ⫽ left side slope; soiuznogo Nauchno-Issledovatel’skogo Instituta Gidrotekhniki 共Trans.
All-Union Sci. Res. Inst. Hydraulic Eng.兲, Leningrad, Vol. 9, pp. 238–
Z2共H兲:1共V兲 ⫽ right side slope;
241.
␭1i, ␭2i ⫽ estimate of the Lagrangian multipliers at
Monadjemi, P. 共1994兲. “General formulation of best hydraulic channel
current iterate;
section.” J. Irrig. Drain. Eng., 120共1兲, 27–35.
␭1, ␭2 ⫽ Lagrangian multipliers associated with Murtagh, B. A., and Saunders, M. A. 共1993兲. “MINOS 5.4 user’s guide.”
constraints C1共x兲 and C2共x兲; Tech. Rep. SOL 83-20R, System Optimization Laboratory, Dept. of
␮ ⫽ mean value of X; Operations Res., Stanford Univ., Stanford, Calif.
␳ ⫽ scalar penalty parameter; Tung, Y. K., and Mays, L. W. 共1980兲. “Risk analysis for hydraulic de-
␴ ⫽ variance SX in sample statistic; and sign.” J. Hydr. Div., 106共5兲, 893–913.
␴2 ⫽ variance of X. Yen, B. C. 共2002兲. “Open channel flow resistance.” J. Hydraul. Eng.,
128共1兲, 20–39.

60 / JOURNAL OF IRRIGATION AND DRAINAGE ENGINEERING © ASCE / JANUARY/FEBRUARY 2007

Downloaded 19 Jan 2009 to 59.162.23.221. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright