Você está na página 1de 22

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225719562

Design of Barrages with Genetic Algorithm Based Embedded Simulation


Optimization Approach

Article  in  Water Resources Management · September 2011


DOI: 10.1007/s11269-010-9706-9

CITATIONS READS

14 780

1 author:

Raj Mohan Singh


Motilal Nehru National Institute of Technology
40 PUBLICATIONS   361 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

http://www.sciencedirect.com/science/article/pii/S0956053X16303014 View project

HYDROLOGICAL MODELING INCLUDING CLIMATE CHANGE USING ArcSWAT View project

All content following this page was uploaded by Raj Mohan Singh on 16 March 2016.

The user has requested enhancement of the downloaded file.


Water Resour Manage (2011) 25:409–429
DOI 10.1007/s11269-010-9706-9

Design of Barrages with Genetic Algorithm Based


Embedded Simulation Optimization Approach

Raj Mohan Singh

Received: 22 May 2009 / Accepted: 10 September 2010 /


Published online: 23 September 2010
© Springer Science+Business Media B.V. 2010

Abstract Barrages are hydraulic structures constructed across rivers to divert flow
into irrigation canals or power generation channels. The most of these structures
are founded on permeable foundation. The optimum cost of these structures is
nonlinear function of factors that cause the seepage forces under the structure. There
is, however, no procedure to ascertain the basic barrage parameters such as depth of
sheet piles or cutoffs and the length and thickness of floor in a cost–effective manner.
In this paper, a nonlinear optimization formulation (NLOF), which consists of an
objective function of minimizing total cost, is solved using genetic algorithm (GA).
The mathematical model that represents the subsurface flow is embedded in the
NLOF. The applicability of the approach has been illustrated with a typical example
of barrage profile. The results obtained in this study shows drastic cost savings when
the proposed NLOF is solved using GA than that of using classical optimization
technique and conventional method. A parametric analysis has also been performed
to study the effect of varying soil and hydrological conditions on design parameters
and on over all cost.

Keywords Subsurface flow · Hydraulic structures · Barrages ·


Embedded simulation optimization · Seepage head ·
Genetic algorithm based optimization

1 Introduction

Rivers of the Indian subcontinent observe high variability of flow over the year.
Design and construction of diversion head works has been a matter of concern for
hydraulic researchers and field engineers. Barrages are major diversion structures of

R. M. Singh (B)
Department of Civil Engineering, Motilal Nehru National Institute of Technology,
Allahabad 211004, India
e-mail: rajm@mnnit.ac.in
410 R.M. Singh

storage for delivering water for various purposes such as irrigation, water supply,
navigation etc. They store large quantity of water making these projects extremely
costly, worth millions of rupees (Rs.). The cost function is nonlinear, consisting of a
large number of design variables of physical dimensions like lengths, thicknesses as
well as unit cost of concrete and earth work. Optimal design of the barrages is one
of the important issues along with water storage, diversions and flood mitigation.
implements cost optimization through a complex multivariate objective function,
deterministic approach is adopted. Though present work is limited to economic
optimization, the inclusion of terms representing the local population and stake
holder’s needs in the objective function may be a correct and realistic objective func-
tion, with all costs not necessarily expressible in terms of monetary value (Bernier
2003).
Barrage is a gate controlled weir. A barrage and weir are similar structures but
differ in qualitative sense. Barrages offer better control than weirs. However, bar-
rages are costlier than weirs (Modi 1988). The characteristics of surface and subsur-
face flows are taken into considerations while designing a barrage. The crest level,
downstream floor length, and minimum depths of upstream and downstream sheet-
piles/cutoffs are mainly governed by surface flow considerations. However, for a
given surface flow condition the cost of a barrage largely depends upon the depth
of sheet-piles and the length of the floor and its thickness which is governed by
subsurface flow conditions. In fact, change in depth of sheet pile affect the floor
length and uplift pressure distribution beneath the floor, and hence thickness of the
floor. Thus, cost function has nonlinear variation with the variation in depth of sheet
piles. The exit gradient, which is considered the most appropriate criterion to ensure
safety against piping (Khosla et al. 1936) on permeable foundations, exhibits non
linear variation in floor length with variation in depth of down stream sheet pile.
These facts complicate the problem and increase the non linearity of the problem.
However, an optimization problem may be formulated to obtain the optimum
structural dimensions that minimize the cost as well as satisfy the exit gradient
criteria.
Optimization methods have been proved of much importance when used with
simulation modeling and the two approaches when combined give the best results
(Rani and Moreira 2010). The optimization problem for determining an optimal
section for the weirs or barrages normally consists of minimizing the construction
cost, earth work, cost of sheet piling, length of impervious floor etc. (Garg et al.
2002; Swamee et al. 1996). This study makes an attempt to formulate a nonlinear
optimization formulation (NLOF) that minimizes unit cost of concrete work, and
earthwork and searches the barrage dimension satisfying the exit gradient criteria.
The subsurface flow simulation is embedded in the optimization model. The NLOF
is solved using Genetic algorithm (GA). The cost of reinforcement (steel work) is
excluded in this study.
In this paper seepage head is varied within a specified range to incorporate
the flood event in the design deterministically. However, inclusions of probabilistic
analysis like mitigation against flood (Jordaan 2005; Eckert et al. 2008, 2009) and risk
cost optimization (Rasekh et al. 2010) in the design may be better option. However,
it is much more difficult to implement probabilistic analysis with a multivariate
objective function (Amzal et al. 2006) such as discussed in this paper.
Genetic Algorithm Based Embedded Simulation Optimization Approach 411

2 Genetic Algorithm

GA is based on the principles of genetics and natural selection and was originally
proposed by Holland (1975) and further developed by Goldberg (1989). Genetic
algorithms are applicable to a variety of optimization problems that are not well
suited for standard optimization algorithms, including problems in which the ob-
jective function is discontinuous, non-differentiable, stochastic, or highly nonlinear
(Haestad et al. 2003). GA manipulates a string of numbers in a manner similar to
how chromosomes are changed in biological evolution. The GA search starts from a
population of many points, rather than starting from just one point. This parallelism
means that the search will not be trapped on local optima (Singh and Datta 2006).
The description of basics of GA is available in literature (Goldberg 1989; De Jong
1975; Passino 2005). The present work utilized a real coded genetic algorithm by
Passino (2005).
Two optimization algorithms certainly differ in convergence speed. When the
problem is non-linear (either objective function or constraints) or multi-modal (hav-
ing more than one peaks or valleys) it is not likely that two algorithm will converge to
same solution. Different optimization algorithm applied to same problem some may
converge to local optima while better one will able to get the global optima. Tra-
ditional classical optimization methods start the search from a single starting point
there for it is likely to stuck into local optima. Genetic algorithm starts the search
from population of points therefore it is likely to get global optima.

3 Mathematical Model of Subsurface Flow Under Barrages

The seepage (water seeping through soil) underneath hydraulic structure may be
represented by:
∂ 2h ∂ 2h ∂ 2h
+ 2 + 2 =0 (1)
∂ x2 ∂y ∂z
This is well known Laplace equation for seepage of water through porous media.
This equation implicitly assumes that (i) the soil is homogeneous and isotropic;
(ii) the voids are completely filled with water; (iii) no consolidation or expansion
of soil takes place; and (iv) flow is steady and obeys Darcy’s law. The subsurface
flow under barrages/weirs will mainly be two dimensional, as the width of a river is
so considerable that the subsurface flow at any cross section of the barrage is not
appreciably influenced by any cross-flow from the sides except near the flanks. For
2-dimensional flow, the seepage Eq. 5 may be written as:
∂ 2h ∂ 2h
+ 2 =0 (2)
∂ x2 ∂y

3.1 Seepage Analysis

The need to provide adequate resistance to seepage flow represented by Eq. 6 both
under and around a hydraulic structure may be an important determinant of its
412 R.M. Singh

Fig. 1 Typical layout of a


barrage in a diversion head
works

geometry (Skutch 1997). The boundary between hydraulic structural surface and
foundation soil represents a potential plane of failure. Uncontrolled seepage may
result in ‘piping’ through soil subjected to an excessive overall hydraulic pressure
gradient that causes soil particles to be dislodged from the matrix, or ‘boiling’ of the
subsoil at the exit caused by a local excess pressure gradient. Figure 1 is a schematic
of a barrage in a diversion headworks subjected to seepage.
Stability under a given hydraulic head could in theory be achieved by an almost
limitless combination of vertical and horizontal contact surfaces below the structure
provided that the total length of the resultant seepage path were adequately long
for that head (Leliavsky 1979; Skutch 1997). In practical terms, the designer must
decide on an appropriate balance between the length of the horizontal and vertical
elements. Design practice prevailing in a particular country or region may effectively
dictate the balance. Optimization based on knowledge of local construction costs
would be possible but is rarely attempted. This work is one of such attempts.
For large structures it may be practicable to conduct seepage analysis using:
(i) Flow nets constructed by trial and error or graphical methods; (ii) Electrical
analogues; (iii) Mathematical solutions of the Laplace Eq. 6. Many researchers
(Parsons 1929; Terzaghi 1929; Khosla 1930, 1932; Harza 1935; Lane 1935; Garg et al.
2002) stated that problem of steady-state subsurface flow can be represented by the
Laplace equation. However, simplified and empirical method presented by Bligh’s
creep theory and Lane theory (Varshney et al. 1988) are commonly used in the rou-
tine design of low to medium head structures. Lane’s Weighted Creep Theory, and
Khosla’s Method of Independent Variables, is most commonly adopted (Varshney
et al. 1988) methods.
Khosla et al. (1936) used method of independent variables based on Schwarz-
Christoffel transformation to solve the Laplace Eq. 6 which represent seepage
through the subsurface media under a hydraulic structure. A composite structure
is split up into a number of simple standard forms (Fig. 3) each of which has a known
solution. The uplift pressures at key points corresponding to each elementary form
are calculated on the assumption that each form exists independently. These key
points are junction points of sheet piles with the impervious concrete floor. The var-
ious forms are superposed and corrections to the pressures are made to allow for the
interaction of each form with the others. An explicit check is made for the stability
Genetic Algorithm Based Embedded Simulation Optimization Approach 413

Fig. 2 Standard profiles: a


sheet-pile at upstream, b
sheet-pile at downstream, c
sheet-pile at any intermediate
point

of the soil at the exit from the structure (exit gradient). Appropriate factors of safety
to ensure stability in different types of soil are incorporated in the safe exit gradients
determined by Khosla et al. (1936). The residual heads between key points are as-
sumed to vary linearly. From literature it has been established that solution based on
analytical solution by Khosla’s with various corrections for actual profiles are com-
parable with that of numerical solution based on finite element methods (Garg et al.
2002).
Solutions of these simple profiles have been obtained in terms of proportion of
head H at key points represented by . The proportion of head at any key point C
is thus HC /H. Here, H is the seepage head (difference between head water and tail
water elevations), and HC is the residual head at C and represent uplift pressure head
at C. Solutions of the standard forms as shown in Fig. 2 are discussed as follows:
For location of sheet pile (cutoff wall) at upstream end,
 
1 2−λ
φC = cos−1 (3)
π λ

 
1 1−λ
φ D1 = cos−1 (4)
π λ
 √ 
where λ = 12 1 + 1 + α 2 ; and α = dL1
For location of sheet pile at downstream end,
 
1 λ−2
φ E = cos−1 (5)
π λ

 
1 λ−1
φ D2 = cos−1 (6)
π λ
 √ 
where λ = 1
2
1 + 1 + α 2 ; and α = L
d2
414 R.M. Singh

For location of sheet pile anywhere between upstream and down stream
 
1 −1 λ1 − 1
φC1 = cos (7)
π λ
 
1 λ1 + 1
φ E1 = cos−1 (8)
π λ
 
λ1
1
φD = cos−1 (9)
π
λ
     
where λ1 = 12 1 + α1 2 − 1 + α2 2 ; λ2 = 12 1 + α1 2 + 1 + α2 2 ; and α1 = bd ;
α2 = L−b
d

3.2 Corrections for Floor Thickness

In the standard forms (Fig. 2), the thickness of floor is assumed to be negligible; hence
the values at key points obtained from Khosla’s equation refer to the top level of
the floor. Corrections have to be applied for thickness of floor, and are interpolated
assuming linear variations along the sheet-pile depth. For example, the thickness
correction at key point C (TC ) (Fig. 2a) will be additive and is given by
 
φ D1 − φC
TC = tmin (10)
d1
where tmin is thickness of concrete floor at key point C.

3.3 Correction for Mutual Interference of Sheet Piles

The amount of correction in percent is given as:


  
D D+d
c = 0.19 (11)
b L
where c is correction to be applied; b  distance between two sheet piles; L is total
floor length; D is depth of sheet pile whose influence has to be determined on the
neighboring sheet pile of depth ‘d’. D is to be measured below the level at which
interference is desired. The correction is positive if the interfering pile is downstream,
and negative for vice-versa. For the general profile of a barrage as shown in Fig. 3,
the corrected values represented as φC and φE at key points C and E, respectively,
may be written as follows:
    
 φ D1 − φC DC DC + dC
φC = φC + tmin + 0.19 (12)
d1 b L

    
 φ E − φ D21 DE DE + dE
φE = φE − tmin + 0.19 (13)
d2 b L
where b  = L − w, and w = width of sheet pile caps (assumed 1.0 m); DC = h1 −
tmin + d2 ; dC = d1 − tmin ; dE = d2 − t; DE = d1 − h1 − t for d1 = h1 + t, and = 0.0 for
Genetic Algorithm Based Embedded Simulation Optimization Approach 415

h4
3 4 5
2 H
1 y16
l23 l34 6 y36 h7
7
C l12
d1
l14 E
l67 d2
l45 l56
L
Fig. 3 Schematic of a barrage profile and parameters utilized in problem formulation and perfor-
mance evaluation

d1 < h1 + t; tmin = minimum thickness of the concrete floor in the upstream; and t =
actual thickness of the concrete floor in the downstream.
The residual heads between key points C and E are assumed to vary linearly.
Hence, the corrected residual head proportion at any distance ‘x’ from the down-
stream end can be written as:
 
 φC − φ E )
φx = φ E + x (14)
L

3.4 Exit Gradient

An explicit check is for the stability of the hydraulic structure for soil at the exit is
devised by Khosla [2] in the from of exit gradient. The exit gradient for the simple
profile as in Fig. 2b is given by Khosla [2] as follows:
H 1
GE = √ (15)
d2 π λ
 √ 
where λ = 12 1 + 1 + α 2 ; α = dL2 ; L is total length of the floor; and H is the
seepage head.
Equation 19 gives G E equal to infinity for no sheet pile at the downstream side
of the floor. Therefore, it is necessary that a vertical cutoff (sheet pile) be provided
at the downstream end of the floor. To prevent piping, the exit gradient is kept well
below the critical values which depend upon the type of soil.
The uplift pressures and exit gradients are two basic essentials which must be
considered while designing a hydraulic structure. The uplift pressures must be
balanced by a suitable thickness of floor at different points, and the exit gradients by
a suitable depth of pile line at the downstream end of the floor. Any optimal design of
hydraulic structure must incorporate these two in the optimization formulation. The
present work uses GA based optimization formulation incorporating uplift pressure
and exit gradient in the optimization model to fix depth of sheet piles and length
and thickness of floor. The optimization solution thus ensures safe structure with
economy.
416 R.M. Singh

4 Subsurface Flow Embedded Optimal Design Formulation

The objective of this hydraulic structure design optimization model formulation is to


determine the dimensions of diversion barrages in terms of its length, upstream sheet
pile depth and downstream sheet pile depth. The over all objective is to minimize
the total cost of the construction of the barrage. The constraints of the optimization
model is that uplift pressures must be balanced by a suitable thickness of floor at
different points along the floor length of the barrage, and the length of the floor
and the depth of sheet piles be such that the exit gradients is well within the safe exit
gradient (SEG) limit. Upper and lower bounds representing possible ranges of values
can be imposed on the decision variables to reduce the search space.
Steady state subsurface flow is assumed to occur under the conditions when the
gates of the barrage are closed and water is ponded upstream for diversion into the
canal.

4.1 Optimization Model

Minimize fc (L, d1 , d2 ) = c1 (f1 ) + c2 (f2 ) + c3 (f3 ) + c4 (f4 ) − c5 (f5 ) (16)

H
Subject to SEG ≥ √ (17)
d2 H λ

Ll ≤ L ≤ Lu (18)

dl1 ≤ d1 ≤ d1u (19)

dl2 ≤ d2 ≤ d2u (20)

L, d1 , d2 ≥ 0 (21)
where fc is objective function represents total cost of barrage per unit width (Rs/m),
and is function of floor length (L), upstream sheet pile depth (d2 ) and downstream
sheet pile depth (d2 ); f1 is total volume of concrete in the floor per unit width for
a given barrage profile and c1 is cost of concrete floor (Rs/m3 ); f2 is the depth of
upstream sheet pile below the concrete floor and c2 is the cost of upstream sheet
pile including driving (Rs/m2 ); f3 is the depth of downstream sheet pile below the
concrete floor and c3 is the cost of downstream sheet pile including driving (Rs/m2 );
f4 is the volume of soil excavated per unit width for laying concrete floor and c4 is cost
of excavation including dewatering (Rs/m3 ); f5 is the volume of soil required in filling
per unit width and c5 is cost of earth filling (Rs/m3 ); SEG is safe exit gradient for a
given soil formation on which the hydraulic structure isconstructed and  is function

of downstream depth and the length of the floor; λ = 2 1 + 1 + α ; α = dL2 ; L is
1 2

total length of the floor; H is the seepage head; d1 is the upstream sheet pile depth;
Genetic Algorithm Based Embedded Simulation Optimization Approach 417

d2 is downstream sheet pile depth; Ll , d1l , and d2l is lower bound on L, d1 and d2
respectively; Lu , d1u , d2u are upper bound on L, d1 and d2 respectively.
The constraint Eq. 17 may be written as follows after substituting the value of λ:


2 ⎞1/2
 2
H
L − d2 ⎝ 2 −1 − 1⎠ ≥0 (22)
d2 π (SGE)

In the optimization formulation, for a give barrage profile and seepage head H, f1
is computed by estimating thickness at different key locations of the floor using
Khosla’s method, and hence nonlinear function of length of floor (L), upstream sheet
pile depth (d1 ) and downstream sheet pile depth (d2 ). Similarly f4 , and f5 is nonlinear.
The constraint represented by Eq. 17 is also nonlinear function of length of the floor
and downstream sheet pile depth (d2 ). Thus both objective function and constraint
are nonlinear; make the problem in the category of nonlinear optimization program
(NLOP) formulation, which are inherently complex.
The Eq. 22 explicitly incorporates subsurface flow under the barrages. The
functional parameters f1 to f5 are affected by subsurface flow, and hence effect of sub-
surface flow is implicitly incorporated in the optimization model. The optimization
formulation embedded with subsurface flow behavior under the barrages is solved
using genetic algorithm to obtain the optimal design parameters.
For a given geometry of a barrage and seepage head H, the optimization model
functional parameters f1 , f2 , f3 , f4 and f5 may be characterized using the procedure
outlined here. A typical barrage profile shown in Fig. 3 is considered for illustration
of the procedure. Intermediate sheet-piles are not effective in reducing the uplift
pressures and only add to the cost of in reducing the uplift pressures and only add to
the cost of the barrage. In present work, no intermediate sheet piles are considered.

4.2 Characterizing f1

Quantity of concrete depends upon the length and the thickness of the concrete floor.
Quantity of concrete upstream of the ponding by barrage gate i.e. portion between
points 1 and 4 (Fig. 3), only a nominal thickness of concrete is required because the
residual head (uplift pressure) is balanced by the weight of water above the floor. If
minimum thickness is denoted by tmin , the volume of concrete per unit width between
points 1 and 4 is given as:

Q14 = (L − (l45 + l56 + l67 )) tmin (23)

The quantity of concrete downstream of the ponding by gate i.e. portion between
4 and 5 (Fig. 3) is estimated by finding the actual thickness depending upon residual
heads at points 4, 5, 6, and 7. If h4 and h7 represent residual head at point 4 and 7
respectively calculated using Eqs. 16 and 17, as shown in Fig. 3, the residual head at
other points may be calculated using Eq. 18 assuming linear variation. The line shown
as dotted in Fig. 3, thus represent hydraulic gradient line (HGL), the ordinate of
which at any point gives residual head at that point. Thus, h4, h5 , h6 , and h7 represent
residual head at locations 4, 5, 6, and 7. If hx is ordinate of HGL from the top of the
418 R.M. Singh

floor at distance x from the downstream end, then thickness (t) required to balance
may be given by
hx
t= (24)
G−1
where G is the specific gravity of the material of the floor. For concrete, it is taken
as 2.42.
Depending upon the magnitude of the ordinates, the thickness, and hence volume
of the concrete varies. For a particular profile of barrage, optimization model require
quantity of concrete to be estimated. Initially, a minimum thickness is provided
between locations 4 to 7, and volume of concrete corresponding to minimum
thickness is represented as Qmin . Depending upon magnitude of ordinates of HGL at
locations 4, 5, 6 and 7, additional thickness is provided. The volume of concrete after
providing additional thickness over the minimum thickness is calculated separately
between the portions 4 and 5, 5 and 6, and 6 and 7 and represented as Q45 , Q56 ,
and Q67 respectively. The following procedure step by step is utilized to estimate the
quantity of concrete between points 4 to 7. The procedure is similar to Garg et al.
(2002).

(i) The volume of concrete per unit width for the minimum thickness is esti-
mated as:

Qmin = (l45 + l56 + l67 ) tmin (25)

(ii) Consider portion between location 4 and 5. Assume HGL ordinate at locations
4 and 5 as h4 and h5 respectively. Depending on the magnitude of h4 and h5,
the quantity of concrete may be calculated as follows:
a. If h4 > 0 and h5 < = 0; then
 
h42
Q45 = 0.5 l45 ; (26)
h4 + |h5 |
b. If h4 > 0 and h5 > 0; then

Q45 = 0.5(h4 + h5 )l45 ; (27)

c. If h4 < = 0 and h5 < = 0; then, Q45 = 0;


(iii) In a similar fashion, consider portion between 5 and 6; and portion between 6
and 7. Depending upon magnitude of h5 , h6 , and h7 , Q56 and Q67 is estimated.
(iv) Total quantity of concrete between portion 4 to 7,

Q47 = Qmin + Q45 + Q56 + Q67 (28)

(v) For small scour depths (assumed up to 3 m), sometimes concrete cutoffs
are provided. In that case quantity of concrete in upstream and downstream
cutoffs represented as Qcutoff(u/s) and Qcutoff(d/s) have to be included. These
quantities estimated as follows:
For d1 < = 3.0 m,

Qcutoff(u/s) = d1 − tmin (29)


Genetic Algorithm Based Embedded Simulation Optimization Approach 419

For d1 > 3.0 m, Q cutoff(u/s) = 0.5


(Assuming cutoff width as 1 m)
Similarly,
For d2 < = 3.0 m,

Qcutoff(d/s) = d2 − t2 (30)

where t2 is actual depth by residual head consideration.


For d1 > 3.0 m, Q cutoff(d/s) = 0.5
(vi) Hence total quantity of concrete per unit width is given as:

QC = Q14 + (Q47 ) / (G − 1) + Qcutoff(u/s) + Qcutoff(d/s) (31)

(vii) Thus,

Q47
f1 = QC = L.tmin + + Qcutoff(u/s) + Qcutoff(d/s) (32)
(G − 1)

4.3 Characterizing f2 and f3

Depending upon the depth of u/s and d/s sheet pile depths that satisfy the subsurface
constraints, these values are supplied to the optimization model.

4.4 Characterizing f4 and f5

The f4 represents quantity of earthwork in excavation. The concrete floor is laid


after excavating the soil on the bed of the river or stream. The volume of earthwork
required to be excavated per unit width for laying the concrete floor consists of (i)
quantity soil equal to the volume concrete in concrete floor; (ii) quantity of soil above
the floor level, and (iii) quantity of soil in the cutoffs.
In Fig. 4, area abb1 a1 represent an elemental in excavation and area acc1 a1
represent elemental area in filling, and represented by functions f(x) and g(x)

b b1
c c1
1 x1 x2 a a1 x3 x4 y1 y
3

l14 7
l67

l45 l56
L
Fig. 4 Schematic of parameters for earth work characterization
420 R.M. Singh

Table 1 Physical parameters Physical parameters Values (meters)


values of barrage profile by
conventional method utilized l12 67.77
for performance evaluation as l23 6.70
shown in Fig. 3 l34 1.0
l45 1.0
l56 9.90
l67 19.0
La 105.37
Y16 1.65
Y36 3.15
H 7.12
a Decision d1a 5.45
variables to be
optimized d2a 5.90

respectively. From the Figs. 3 and 4, the volume of soil excavated per unit width
is estimated by Garg et al. (2002):
(Q45 + Q56 + Q67) y2
f4 = QEarth Excavation = L.tmin + + y1 l67 + 0.5 1 l56
(G − 1) y3
⎛ ⎞
x4 x3
⎜ ⎟
− ⎝ f (x) ∂ x − g (x) ∂ x)⎠ + Qcutoff(u/s) + Qcutoff(d/s) (33)
x1 x2

The volume of soil per unit width in filling is estimated by


x3
f5 = QEarth Filling = g (x) ∂ x (34)
x2

In present work, for performance evaluations, value of cost of concreting, c1 ,is taken
as Rs 986.0/m3; cost of sheet-piling including driving, c1 , is taken as Rs1510.0/m2;
cost of excavation and dewatering, c3 , is taken as Rs 35.60/m3; cost of earth filling,
c4 , ia taken as Rs 11.0/m3; and minimum thickness of floor is 1.0 m by conventional
method. The dimensions of the barrage profile as shown in Fig. 3 are presented in
Table 1.

5 Solution Procedure of Optimization Formulation Using GA

The optimization model represented by Eqs. 16–22 and the functional parame-
ters embedded in the optimization model and represented through Eqs. 23–34 is
solved using Genetic Algorithm on MATLAB platform (Passino 2005). A classical
constrained nonlinear optimization (nonlinear programming) function, FMINCON,
from Optimization Toolbox in MATLAB, is also utilized to compare the results.

5.1 Solution Procedure

The flow chart of the solution procedure of GA employed in this study is depicted
in Fig. 5. As evident from the optimization formulation equations, the specified
Genetic Algorithm Based Embedded Simulation Optimization Approach 421

Fig. 5 Schematic
Specification of parameters (decision variables) of
representation of GA-based problem domain in optimization formulation
optimization model embedded
with seepage simulation
Representation of solution space by string of
beneath the barrages chromosomes of specified lengths where each
individual (chromosomes) correspond to a parameter

For a specified Randomly generate initial population of potential


seepage head, values of parameters in forms of strings
simulate
seepage flow
to find f1, f2, f3, Decode each individual into decimal valued
f4, and f5 parameter
parameters in
optimization Assign fitness value of each individual of
formulation population using objective function and constraints

Optimal Yes
Termination
parameters
criteria met
obtained

No
Select and met the individual with high fitness
value with more fit individual end up with
more copies of themselves

Perform cross-over operation on the selected


parent population

Perform mutation operation as in cross over


operation with low probability

Obtain new population after cross-over and


mutation

variables are length of the weir (L), upstream (u/s) sheet-pile depth (d1 ), and down-
stream (d/s) sheet-pile depth (d2 ). GA starts with representation of these variables
by individual chromosomes of specified number or genes. Each number slot in a
chromosome is called a gene. A six-digit chromosome (1 to 9 digit as one of six genes)
is selected based on Passino [11] guidelines. GA starts with randomly generated
population of solution vector parameters (L, d1 , d2 ). Each of these parameters was
decoded into decimal valued numbers. Employing Khoslas’ method to incorporate
under seepage, f1 , f2 , f3 , f4 , and f5 were estimated. Using objective function value
and constraints, fitness of parameters in whole population were found out. From
this population, new population was created using genetic operators (selection,
crossover, and mutation). Thus, from initial population, next generation of new
population was completed.
The process of going into new generation continues until the fitness of the popu-
lation converged or termination criteria satisfied. Elitism was employed to preserve
and carry over the most fit individual in each generation to the next generation with-
out being modified in any way by the genetic operators. The population converged
at crossover rate of 0.5, mutation rate of 0.05, the population size is 40 and the
422 R.M. Singh

Fig. 6 Average, best and


worst fitness versus number of
generations

number of generations is 100. The average fitness, best fitness and worst fitness in
100 generations are also plotted and shown in Fig. 6.

5.2 Solution by Classical Optimization

The optimization model and the functional parameters embedded in the optimiza-
tion model are solved using non-linear constrained optimization function ‘FMIN-
CON’ of MATLAB (2007). The optimization model represented by Eqs. 16–22 and
the functional parameters embedded in the optimization model and represented
through Eqs. 23–34 and the functional parameters embedded in the optimization
model are solved using non-linear constrained optimization function “FMINCON”
of MATLAB.
The function ‘FMINCON’ finds a constrained minimum of a function of several
variables. ‘FMINCON’ attempts to solve optimization problems of the form:
minF (X) (35)

subject to : A ∗ X <= B, Aeq ∗ X = Beq (linear constraints) (36)

C (X) <= 0, Ceq (X) = 0(nonlinear constraints) (37)

 
LB <= X <= UB lower and upper bounds (38)
In MATLAB the solution may be represented by:
 
X = FMINCON FUN, X0, A, B, Aeq, Beq, LB, UB, NONLCON (39)
subjects the minimization to the constraints defined in NONLCON. The function
NONLCON accepts X and returns the vectors C and Ceq, representing the non-
linear inequalities and equalities respectively. FMINCON minimizes FUN such that
C(X)<= 0 and Ceq(X) = 0.
The basic steps employed in solution procedure may be presented as follows:
(i) Specification of parameters (decision variables) of problem domain in opti-
mization formulation
(ii) For a specified seepage head, simulate seepage flow to find f1 , f2 , f3 , f4 , and f5
parameters in optimization formulation in FUN.
Genetic Algorithm Based Embedded Simulation Optimization Approach 423

Table 2 GA and conventional methodology based results with safe exit gradient equal to 1/8 and
minimum thickness of floor as 1 m
Design method U/S sheet D/S sheet Floor length Cost Cost reduction
pile d1 (m) pile d2 (m) L (m) (Rs/m) by GA (%)
Conventional method 5.45 5.90 105.37 133,605.0
GA 3.0 9.20 61.02 111,250.0 16.73

(iii) Assign constraints in NONLCON.


(iv) Run FMINCON with assumed starting point (X0) and lower and upper bounds
of decision variables.
Both solution methods (GA and Classical optimization) consume very less time (less
than 30 s). The GA run for 100 generations consumes 23.35 s where as classical
optimizations run takes 2 s.

6 Performance Evaluation

Performance of the developed methodology is evaluated by solving a typical barrage


profile. Till the mid nineteenth century in India, weirs and barrages have been
designed and constructed on the basis of experience using the technology available at
that period of time. Some of them were based on Bligh’s creep theory, which proved
to be unsafe and uneconomical. Comparison of the parameters of these structures
with the proposed approach is, thus, not justified. Therefore, a typical profile, a
spillway portion of a barrage as shown in Fig. 3 is chosen for illustrating the proposed
approach using dimensions values of parameters as given in Table 1. The proposed
approach shown in Fig. 5 is implemented in steps as discussed in the previous section
of solution procedure of optimization formulation using genetic algorithm.

7 Results and Discussions

The barrage profile is first designed by the conventional method based on Khosla’s
2-D seepage analysis, in which the depth of sheet-piles is limited from scour consid-
erations and the floor length is established to achieve a permissible exit gradient. The
barrage profile is then optimized using the proposed approach with the same relative
prices of materials used in the conventional method. In the optimization approach
the depth of sheet-piles determined from scour considerations is taken as a lower
bound (3.0 m), and the upper bound is set from practical considerations and limited
to 12.0 m. In present work, for performance evaluations, value of cost of concreting,

Table 3 Optimum cost of Safe exit gradients d1 d2 L Cost (Rs)


barrage by GA at minimum
thickness of floor as 1 m and 1/8 3.0 9.20 61.02 111,250.0
variable safe exit gradients 1/7 9.16 9.81 40.36 104,340.0
1/6 3.0 7.65 40.0 91,408.0
1/5 3.0 5.59 40.0 82,626.0
1/4 3.0 3.74 40.0 73,656.0
424 R.M. Singh

Fig. 7 Variation of optimum 120000


costs with exit gradients
100000

80000

Cost
60000
40000

20000

0
0.13 0.14 0.17 0.20 0.25

Exit gradient

c1 , is taken as Rs. 986.0/m3 ; cost of sheet-piling including driving, c1 , is taken as Rs.


1510.0/m2 ; cost of excavation and dewatering, c3 , is taken as Rs. 35.60/m3 ; cost of
earth filling, c4 , ia taken as Rs. 11.0/m3 ; and minimum thickness of floor is 1.0 m by
conventional method. The results obtained are presented in Table 2.
The results in Table 2 shows a saving of approximately 17% in the barrage cost
by using GA based subsurface flow embedded methodology. The results in Table 2
clearly shows that GA based optimization approach favors deeper depth of down
stream sheet piles than required from scour considerations. It resulted in a smaller
floor length and overall lower cost.
The cost of barrages will not remain the same for different types of soil and
hydrological conditions. The different types of soil and different hydrological con-
ditions are addressed by incorporating exit gradients values and randomly generated
seepage head values in the GA based subsurface flow embedded methodology. A
parametric analysis was performed to illustrate the effects of exit gradients on over
all cost. The parametric analysis is also conducted to show effect of assumed thickness
and seepage head on over all cost of the barrages and average downstream thickness
of the barrage profile.

7.1 Parametric Analysis to Illustrate Effect of Safe Exit Gradient

The safe exit gradient depends upon the type of soil on which barrage is founded.
The SEG for sand is higher than that of silt (lower grain size). The GA based
subsurface flow embedded model is run with various values of SEG, and the results

Fig. 8 Variation of optimum 120000


costs with different minimum
100000
Optimum Costs (Rs)

assumed floor thickness values


80000

60000

40000

20000

0
0.25 0.5 0.75 1
Minimum assumed floor thickness (m)
Genetic Algorithm Based Embedded Simulation Optimization Approach 425

Table 4 GA and classical optimization based results with safe exit gradient equal to1/7 and variable
minimum floor thickness
GA Classical optimization
T d1 d2 L Cost (Rs/m) d1 d2 L Cost (Rs/m) Cost reduction (%)
1.00 9.16 9.81 40.36 104,340.0 5.45 10.42 51.61 111,418.0 6.4
0.75 3.0 6.30 73.3 95,971.0 5.45 8.58 67.55 103,549.0 7.3
0.50 3.0 5.61 83.92 83,755.0 5.45 6.97 91,251.0 87.10 8.2
0.25 3.0 5.0 100.02 67,205.0 5.45 5.90 105.37 72,651.0 7.5
t minimum thickness of floor in meter, d1 U/S sheet pile in meter, d2 D/S sheet pile in meter, L floor
length in meter

are presented in Table 3. It can be noted from Table 3 that as SEG increases for a
particular assumed minimum thickness the cost reduces. Thus a barrage founded on
sand (high SEG) will cost less than that founded on silt or clay. The results are also
plotted in Fig. 7.

7.2 Parametric Analysis to Illustrate Effect of Minimum Assumed Thickness

A nominal floor thickness is required on the upstream side of the barrages as the
uplift pressures on the upstream side are counterbalanced by the weight of standing
water. However, the downstream thickness needs to be evaluated starting with the
nominal thickness for the whole of the floor length. The results of the analysis for
a range of assumed minimum thickness from 1.0 m to 0.25 m are plotted in Fig. 8.
The permissible exit gradient is taken as 1/7 for the considered barrage profile. The
results with relevant details are shown in Table 4. The decrease in assumed nominal
floor thickness resulted in a increase in total floor length and a decrease in depth
of the downstream sheet pile without appreciably affecting the upstream sheet-pile
depth. This resulted in a substantial decrease in the total cost of the barrage, as shown
in Table 4 and plotted in Fig. 8. Figure 9 shows the increase in average d/s thickness
with increase in the assumed nominal floor thickness. However, a minimum possible
thickness from practical considerations should be provided.
Results in Table 4 indicate that cost reduction range using GA varies from 6 to 8%
than those of using classical optimization technique which is significant for usually

Fig. 9 Variation of average 2.5


Average d/s barrage thickness (m)

thickness with different


minimum assumed floor 2
thickness values for SEG equal
to 1/7 1.5

0.5

0
0.25 0.5 0.75 1
Minimum assumed floor thickness (m)
426 R.M. Singh

Fig. 10 Variation of optimum 60.00


costs with seepage head
50.00

Cost (10^4 Rs)


40.00

30.00

20.00

10.00

0.00
6.81 9.74 16.81 25.83 26.49 28.26 33.46 34.42 43.24 46.06
Seepage Head

Fig. 11 Variation of d/s 15.00


average thickness with seepage
Average Thickness (d/s)

head
10.00

5.00

0.00
6.81 9.74 16.81 25.83 26.49 28.26 33.46 34.42 43.24 46.06
Seepage head (m)

Fig. 12 Effect of variation of 12.50


seepage head on d/s sheet pile
D/S Sheet pile depth (m)

depth 12.00

11.50

11.00

10.50

10.00

9.50
6.81 9.74 16.81 25.83 26.49 28.26 33.46 34.42 43.24 46.06
Seepage head (m)

Fig. 13 Effect of variation of 120.00


seepage head on total floor
length 100.00
Floor length (m)

80.00

60.00

40.00

20.00

0.00
6.81 9.74 16.81 25.83 26.49 28.26 33.46 34.42 43.24 46.06
Seepage head (m)
Genetic Algorithm Based Embedded Simulation Optimization Approach 427

high cost of barrages. It is evident from the Table 4 that cost decreases with decrease
in minimum thickness assumptions for whole of the floor length. It also can be noted
that for identical upstream sheet pile depth and slight increase in downstream sheet
pile depth (increase by 0.61 m), even with more than 16% lesser total floor length the
total cost for minimum floor thickness of 0.5 m increased by almost 20% than those
with minimum thickness of 0.25. The cost per unit is of course higher for sheet pile
than that of unit rate of concrete. The increase in cost is mainly due to increase in
average thickness of d/s length of barrage.

7.3 Parametric Analysis to Illustrate Effect of Seepage Head

Parametric analyses were also carried out with randomly generated different values
of seepage head, with a fixed value of permissible exit gradient of 1/8. The results of
model runs are plotted in Figs. 10, 11, 12, and 13. In general, the results show that
the cost of barrage profile and average downstream barrage floor thickness increases
with increase in seepage head as shown in Figs. 10 and 11. It has been found that there
is no definite trend about the increase in upstream sheet pile depth with seepage
head. Thus, the upstream sheet-pile (cutoff wall) depth is not sensitive to a change in
seepage head.
However, there is increase in down stream sheet pile depth with increase in
seepage head until it approaches its upper bound as shown in Fig. 12. Thereafter, only
the floor length and/or average thickness of downstream floor increases (Figs. 11, 12)
to satisfy the permissible exit gradient, resulting in a sharp increase in the total cost of
the barrage as evident from Fig. 10. The irregularities in the curves of Figs. 10–13 are
effect of the non linearity of the objective function and/or constraints with possible
compensations. There is non linear relation between the length of the impervious
floor, seepage head and depth of downstream sheet pile (Eq. 22). Also, descritization
steps of seepage head is not uniform as different values of seepage head is obtained
by generating random numbers (uniform distribution) within a specified range
in this study.

8 Conclusions

The illustrated application establishes the potential applicability of the genetic algo-
rithm based subsurface flow embedded methodology. The GA based optimization
model is embedded with the subsurface flow simulation to solve the nonlinear objec-
tive function of minimizing cost subject to nonlinear constraints. The results obtained
by solution of the GA based optimization formulation provide the optimal barrage
parameters in terms of depth of up streams and down streams sheet-piles/cutoffs,
length and thickness of floor, and reduces the over all costs. The optimization
approach is capable of evolving a cost effective design of a barrage than that obtained
by conventional method.
A parametric analysis was performed to ascertain relative sensitiveness of barrage
parameters with different soil types and different hydrological conditions. Soil types
and hydrological conditions are incorporated in the model by safe exit gradient and
seepage heads respectively. Safe exit gradients and seepage heads are varied within
428 R.M. Singh

a wide range to represent favorable as well as adverse design scenarios. The conclu-
sions from this study may be summarized as follows:

1. A barrage founded on silt or silty clay is costlier than those of founded on sandy
soils for the same head.
2. The depth of the upstream sheet pile is the least sensitive to the value of SGE or
the seepage head compared to downstream sheet pile and/or total floor length.
3. The minimum assumed floor thickness affects design parameters, and hence
overall cost of the barrage. A nominal value from practical considerations is
imperative to reduce the over all cost of the barrage.
4. As the seepage head increases, the depth of the downstream sheet-pile and floor
length both increase. When the downstream sheet-pile attains to its upper bound,
the optimal floor length increases rapidly to satisfy the SGE, and results in rapid
increase in cost. Too large downstream sheet-pile may result in excessive ponding
at the barrage. Therefore, upper bounds on sheet pile depth should be set to
restrict the extent of ponding at the barrage, and at the same time it should not
excessively increase the over all cost.
5. GA based subsurface flow embedded model is able to evolve more cost effective
design of barrage profile than those of classical optimization technique for the
same problem.

The limited performance evaluation results show the potential applicability of the
GA based methodology for optimizing the barrage profiles dimensions to obtain
optimal costs. Evaluation of the developed methodology on larger problems consid-
ering for both surface and subsurface flows will further increase the applicability of
the methodology.
The study suggests that the GA based optimization methodology as described in
this study is potentially capable to develop a cost–effective design of a barrage. The
GA based optimization approach is equally valid for optimal design of other major
hydraulic structures, such as canal drops and regulators.

References

Amzal B, Bois FY, Parent E, Robert CP (2006) Bayesian-optimal design via interacting particle
systems. J Am Stat Assoc 101(474):773–785
Bernier J (2003) Décisions et comportements des décideurs face au risque. Hydrol Sci J 48(3):
301–316
De Jong KA (1975) An analysis of the behavior of a class of genetic adaptive systems. Ph.D.
dissertation, Univ. of Mich., Ann Arbor
Eckert N, Parent E, Faug T, Naaim M (2008) Optimal design under uncertainty of a passive defense
structure against snow avalanches: from a general Bayesian framework to a simple analytical
model. Nat Hazards Earth Syst Sci 8:1067–1081
Eckert N, Parent E, Faug T, Naaim M (2009) Bayesian optimal design of an avalanche dam using a
multivariate numerical avalanche model. Stoch Environ Res Risk Assess 23:1123–1141
Garg NK, Bhagat SK, Asthana BN (2002) Optimal barrage design based on subsurface flow consid-
erations. J Irrig Drain Eng 128(4):253–263
Goldberg DE (1989) Genetic algorithms in search, optimization and machine learning. Kluwer
Academic Publishers, Boston
Haestad M, Walski TM, Chase DV, Savic DA, Grayman W, Beckwith S, Koelle E (2003) Advanced
water distribution modeling and management. Haestad Press, Waterbury, pp 673–677
Harza LF (1935) Uplift and seepage under dams on sand. Paper No. 1920, Trans. ASCE
Genetic Algorithm Based Embedded Simulation Optimization Approach 429

Holland JH (1975) Adaptation in natural and artificial systems. University of Michigan Press, Ann
Arbor
Jordaan I (2005) Decisions under uncertainty. Pobabilistic Analysis for Engineering Decisions.
Cambridge University Press
Khosla AN (1930) Stability of weirs and canal works: an application of the new theory of hydraulic
gradient. Paper No. 140, Punjab Engineering Congress, Punjab, India
Khosla AN (1932) Pressure pipe observations at Panjnad Weir. Paper No. 160, Punjab Engineering
Congress, Punjab, India
Khosla AN, Bose NK, Taylor EM (1936) Design of weirs on permeable foundations. CBIP Publica-
tion No. 12, Central Board of Irrigation and Power, New Delhi
Lane EW (1935) Security from under seepage. Paper No. 1919, Trans ASCE
Leliavsky S (1979) Irrigation engineering: canals and barrages. Oxford and IBH, New Delhi
Modi PN (1988) Irrigation water resources and water power engineering. Standard Book House,
New Delhi
Parsons HD (1929) Hydraulic uplift in previous soils. Paper No.1713, Trans. ASCE
Passino KM (2005) Biomimicry for optimization, control, and automation. Springer, London
Rani D, Moreira MM (2010) Simulation–optimization modeling: a survey and potential application
in reservoir systems operation. Water Resour Manage 24:1107–1138
Rasekh A, Afshar A, Afshar MH (2010) Risk-cost optimization of hydraulic structures: methodology
and case study. Water Resour Manag. doi:10.1007/s11269-010-9582-3
Singh RM, Datta B (2006) Identification of unknown groundwater pollution sources using genetic
algorithm based linked simulation optimization approach. J Hydrol Eng 11(2):101–109
Skutch J (1997) Minor irrigation design DROP—design manual hydraulic analysis and design of
energy-dissipating structures. TDR Project R 5830, Report OD/TN 86
Swamee PK, Mishra GC, Salem Adel AS (1996) Optimal design of sloping weir. J Irrig Drain Eng
122(4):248–255
Terzaghi C (1929) Effect of minor geological details on the safety of dams. Tech. Publication No.
215. Institute of Mining and Metallurgical Engineers, US
Varshney RS, Gupta SC, Gupta RL (1988) Theory and design of irrigation structures. Nem Chand
and Bros., Roorkee

View publication stats

Você também pode gostar