You are on page 1of 13

Neural Computing and Applications (2022) 34:20313–20325

https://doi.org/10.1007/s00521-022-07591-w (0123456789().,-volV)(0123456789().,-volV)

ORIGINAL ARTICLE

Efficiency of the evolutionary methods on the optimal design of secant


pile retaining systems in a deep excavation
F. Taiyari1 • M. Hajihassani2 • M. Kharghani3

Received: 30 September 2020 / Accepted: 29 June 2022 / Published online: 22 July 2022
 The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2022

Abstract
Deep large excavations in urban areas are an important engineering challenge, whereas secant piling techniques are among
the best solutions to have a safe workplace environment. Optimal design of these structures will increase efficiency as well
as reduce costs. In this paper, the optimum design of secant pile walls as a retaining system of a deep excavation pit is
evaluated. For this purpose, an on-going Tabriz metro station project is investigated as the case study. The structural piles
are made of steel material with a hollow pipe section. A layer of struts is also considered for the horizontal bracing of the
excavation pit. A detailed finite element model is developed in the OpenSees platform in order to perform static analyses.
The optimization of the retaining system is conducted by the mean of four different metaheuristic algorithms including
genetic, particle swarm optimization, bee, and biogeography-based optimization algorithms. The total cost of retaining
structures is considered as an objective function, which should be minimized in the design space of the variables. The
results highlight the excellence of the bees algorithm in achieving a minimum cost, lower dispersion, and rapid conver-
gence rate. The optimum placement of the bracing system and its effect on the soil shear stress are also investigated based
on the obtained optimal results.

Keywords Metaheuristic optimization technique  Retaining system  Steel piles  Deep excavation  Static analysis 
Optimal design

1 Introduction employed for the construction of different components


such as railway stations.
As the population of the world increases, controlling the The stability of deep excavations under the external
facilities and the way of the utilization of space become applied forces such as the lateral earth pressure, surcharge
more critical. As an alternative, deep excavation may be load, hydrostatic pressure, and earthquake loads can be
achieved by some structural members called retaining
systems. The cantilever retaining walls, contiguous bored
piles, secant piles, and sheet piling are some examples of
retaining systems that are frequently used in the excavation
& M. Hajihassani
m.hajihassani@urmia.ac.ir pits. The application of the proper retaining structures
depends on the soil condition, economy of the project, and
F. Taiyari
f-taiyari@tvu.ac.ir the possibility of the implementation of the system in the
project location [1]. The advantages and the disadvantages
M. Kharghani
Kharghani.m@gmail.com of different retaining systems are fully discussed by
Muthomi [2].
1
Department of Civil Engineering, Technical and Vocational Design and performance of supporting systems under
University (TVU), Tehran 1435761137, Iran both static and dynamic loading conditions are studied by
2
Department of Mining Engineering, Faculty of Engineering, several researchers lately [3–8]. Feng et al. [9] used an
Urmia University, Urmia, Iran innovative pre-stressed support (IPS) to restrict the lateral
3
Faculty of Engineering, Islamic Azad University, Science and displacement of the excavation pit. Vahedifard et al. [10]
Research Branch of Tehran, Tehran, Iran

123
20314 Neural Computing and Applications (2022) 34:20313–20325

investigated the effects of negative pore pressure on the solution. Therefore, it is necessary to adopt different
active soil pressure and presented an approximate method metaheuristic algorithms to a specific problem [42].
for involving it during the design process. The reliability Aydogdu [43] used a new version of biogeography-
index of the cantilever retaining structures has been studied based optimization technique with levy flight distribution
by Chakraboty et al. [4]. The behaviour of the reinforced (LFBBO) to optimize reinforced concrete cantilever
soil retaining walls under horizontal and vertical seismic retaining walls under seismic loading. The influence of the
loadings is investigated by Fan et al. [8]. It was demon- seismic load on the optimal cost of the wall was also
strated that the vertical ground acceleration may change the investigated in the above-mentioned reference. Ceranic
shear wave propagation in the reinforced soil and mostly et al. [32] and Yepes et al. [33] optimized a reinforced
increase the reinforcement load. concrete cantilever retaining structure by applying the
Although deep excavations and their supporting struc- simulated annealing optimization algorithm. The outputs of
tures are mostly temporary; however, they are classified as the studies are satisfying and highlight the efficiency of the
the high-risk projects in the engineering field. The list of applied algorithms in minimizing the retaining walls cost.
catastrophic failure of the retaining systems during the last The optimization of the safety factors of retaining walls
decades focused concentrations on providing a reliable conducted by Gordan et al. [31] through network and bee
design guideline [11–13]. Peck [11] developed a simple colony techniques under both static and dynamic condi-
method based on dimensionless curves to estimate the site tions. Mergos et al. [32] and Koopialipoor et al. [33]
settlement near the excavation pit which is widely accepted applied novel optimization techniques for designing the
by engineers. Poulos and Chen [12] presented the two- retaining walls and compared their performance with some
stage analysis method to determine bending moments and other algorithms. Kaveh et al. [34] used eleven meta-
deflections of a pile. Despite the simplicity and the appli- heuristic algorithms in order to obtain the optimum design
cability of the mentioned design methods, they are not of concrete cantilever retaining walls under both static and
suitable in the economical point of view [14, 15]. For dynamic loadings. They applied the Rankine and the
instance, according to the observations discussed by Xu Coulomb methods to determine the lateral earth pressure
and Hatami [15], the reinforced soil retaining walls have and compared the performance of the utilized algorithms.
demonstrated great seismic performance with respect to the Dahake and Saraf [35] solved the same optimization
traditional gravity retaining systems. However, the typical problem by the aid of relief shelve. They showed the
design approach based on the pseudo-static method is the excellence performance of this algorithm on reducing the
same for all systems and does not consider the specific lateral earth pressure. Garcia et al. [36] applied a k-means
details of each system. Therefore, the great potential of the cuckoo search hybrid algorithm to optimize the perfor-
reinforced retaining walls cannot be accounted in their mance of counterfort retaining walls and compared the
design [15]. results with the corresponding outputs based on the har-
Since the retaining systems are usually multi-parameter mony search.
problems, the application of the classical techniques may Due to the availability of the design guideline of the
not result in an optimum solution in most of the time. In cantilever retaining systems in most of the design codes
contrary, the optimum design approaches based on the which provides a straightforward formulation of these
metaheuristic techniques in the geotechnical engineering as systems, almost all the optimization researches are focused
well as the retaining structures are very successful in on these types of retaining walls. In addition, study of the
estimating the optimal solutions. other related systems requires a detailed finite element
In the past few years, various artificial intelligence model which makes its study harder. Therefore, there is
techniques have extensively been utilized for advanced lack of information dealing with other retaining structures.
analysis in a diversity of science and engineering applica- One of the most popular retaining systems which is
tions [16–28]. These techniques have been used along with mostly used for the deep excavations is secant pile walls.
optimization methods for advanced analysis in a diversity These systems comprised of two different parts of struc-
of civil engineering applications, and the use of them is tural and non-structural piles. The structural piles are made
increasing [29–31]. Several studies have also been con- of reinforced concrete or steel material and designed to
ducted on the optimum design of the retaining systems resist the lateral exerted forces, while the non-structural
based on the various metaheuristic optimization algorithms piles are usually made of cement-bentonite–sand slurries
[32–51]. These optimization techniques are able to solve and transfer the soil and/or ground water pressure to the
complicate problems by searching a large possible design closely adjacent structural piles.
space. However, due to their stochastic form, the final In this paper, the complementary evaluations for con-
outcome cannot be guaranteed as the global optimal sidering the effects of pile materials are investigated. The
temporary excavation pit of the on-going Tabriz metro

123
Neural Computing and Applications (2022) 34:20313–20325 20315

station project is selected for this purpose. The piles velocity and displacement terms [54]. These terms can be
dimensions (pipe thickness and outer diameter) together updated in every iteration, according to Eqs. (1–2).
with six extra parameters are selected as design variables
X tþ1
i ¼ X ti þ V tþ1
i ð1Þ
which should be minimized in the design space. The secant
pile wall and the step-by-step excavation process are sim- V tþ1
i ¼ xV ti þ C 1 r 1 ðPi  X ti Þ þ C 2 r 2 ðPg  X ti Þ ð2Þ
ulated in a finite element software. Subsequently, opti-
where X ti and V ti are the current position and velocity of ith
mization process is performed by the use of some
particle, and X tþ1i , and V i
tþ1
are its updated position and
metaheuristic optimization algorithms including genetic,
particle swarm optimization, bees, and biogeography-based velocity, respectively. Pi , and Pg are the best local and
optimization algorithms. At the end, the best-obtained global positions, respectively; x is the inertia weight within
solution of four considered optimization techniques is [0,1.2]; C 1 , and C 2 are random weighting values within [0,
presented, and the efficiency of the algorithms in reducing 2]; r 1 , and r 2 are stochastic values within [0,1]. According
the total cost of the system is fully discussed. to Eq. (2), there are three different contributions in the
movement of each particles. The first term is the previous
motion of the particle, the second term is the particles
2 Metaheuristic optimization techniques cognition, and the third one is their social learning.

2.1 Genetic algorithm 2.3 Bees algorithm

The genetic algorithm (GA) is a stochastic optimization The bees algorithm (BA) is another metaheuristic algo-
technique which follows the principle of natural competi- rithm which has both local and global search capability
tion of individuals in order to appropriate limited sources. using exploitation and exploration strategies, respectively
It is clear that the rate of success generally depends on the [55]. This algorithm was developed by Pham et al. [56] and
genes of the winners. Therefore, reproducing the winner models the foraging behaviour of honey bees to minimize
individuals results in spreading their genes. The selection the objective function.
and reproduction of superior individuals lead to achieve In the BA, n scout bees are randomly sent to selected
more natural resources. Cross-over and mutation are two sites. The fitness functions of the locations (sites) are
key factors which are used to reproduce stronger evaluated and sorted. The m top locations are divided into
individuals. two group of elite and non-elite sites. The neighbourhood
The GA computes the optimum objective function by of the best sites (elite sites) is locally searched by recruiting
simulating this process. In this algorithm, an initial popu- some forager bees. In addition, a random search process is
lation is generated in the search space and evolved towards done for the (n-m) non-best sites as a global search process.
best solution. The individuals have specific genes repre- At the end, the overall sites are sorted based on their
senting their capability during the competition. These finesses.
genes can be altered by the use of cross-over or mutation.
The process is done for some iterations with a new gen- 2.4 Biogeography-based optimization
eration, and its objective function is evaluated at each
iteration. The iteration process is terminated when either a The biogeography-based optimization (BBO) algorithm is
maximum number of populations is generated, or a satis- a recently developed metaheuristic algorithm which was
fying result is reached. The initial form of the algorithm introduced by Simon [57]. The algorithm is inspired by the
was first introduced by John Holland in 1960 based on the biogeography science which simulate the geographical
well-known evolution theory of Darwin. Then, it was distribution of biological organisms in nature.
extended to more practical type by Goldberg [52]. BBO models the migration and the extinction of a
species between islands based on probabilistic mathemat-
2.2 Particle swarm optimization ical relations. In fact, each habitat in the algorithm is a
solution candidate for the optimization problem, and the
The particle swarm optimization (PSO) algorithm is one of position of it in the search space is represented by suit-
the most well-known metaheuristics which was proposed ability index variable (SIVs). The quality of the habitats is
by Kennedy and Elbert [53]. PSO is inspired from social estimated by the habitat suitability index (HSI) which is
behaviour of bird flocking and schooling in nature. In the proportional to its fitness. Accordingly, habitats with higher
standard PSO, a number of particles randomly scattered HSI include better living standards for the species (i.e.
into the search space and generate an initial population. indicates a better solution); therefore, immigration and
The location of each particle is estimated by their specific

123
20316 Neural Computing and Applications (2022) 34:20313–20325

emigration rates of these habitats are low and high, structural steel piles to a depth of (22 ? X5) m on both
respectively. sides of excavation pit (X5 is the required buried height of
The BBO algorithm consists of two key operators, i.e. the piles) and will be continued by the 22 step excavations.
the migration and the mutation. According to the first A layer of soil with 1 m thickness is extracted in each
operator, a new solution is created by some modification in excavation step until the X4 level. Then, a layer of struts is
the design variables of the current solution. The modifi- perpendicularly connected to the pile system to avoid it
cation probability depends on the immigration rate of that from further lateral deformation. This is also illustrated in
solution. The modification is done by using the roulette Fig. 2.
wheel selection method which is directly related to the A thorough soil investigation was conducted for the
probability of the emigration. According to the second studied project. Table 1 summarizes the important infor-
operator, the population in the islands is increased. mation of the soil properties, which are required for the
Therefore, the new solution is created randomly. numerical modelling. It is clear that the soil parameters are
generally on the conservative side. Additionally, the water
table is located 12 m below the ground surface.
3 Case study

Figure 1 demonstrates the place and surrounding space of 4 Numerical simulation


an on-going Tabriz metro station (highlighted in blue). The
station is positioned on the North East side of Tabriz, Iran. The simulation of the pile wall retaining system during the
It is in a crowded residential area and is classified as a different excavation stages is conducted by using OpenSees
high-risk project. The overall length and width of the platform. For sake of simplicity, a two-dimensional model
excavation pit are 70 and 25 m, respectively, and the with the soil domain of (22 ? X5) m tall and 50 m wide
railway line is going to pass through the station at about with the maximum allowable soil element size of 0.5 m is
22 m underground. One of the possible scenarios for the created. Because of the symmetry, only half of the whole
retaining system of the excavation sides is the secant pile span is modelled in the software. The lateral extents of the
system together with a layer of struts, which is going to be soil domain were set to reduce the effects of the boundaries
investigated in this paper. The project applies the cut-and- on the performance of the pile–soil system. The out of
cover technology. plane thickness of the soil elements is X3 (see Fig. 2).
The details of the structural and non-structural pile Figure 3 illustrates the archetype plane strain model of the
arrangements are given in Fig. 2. The implementation of excavation. This figure represents the interface modelling
the considered retaining system is started by driving the

Fig. 1 Location of the project

123
Neural Computing and Applications (2022) 34:20313–20325 20317

Fig. 2 The applied secant pile


retaining system for the
considered excavation pit

Table 1 Soil properties of the case study area


Layer Soil Thickness Weight density (kN/ Cohesion Friction angle Shear modulus Bulk modulus
group (m) m3) (kPa) () (MPa) (MPa)

1 CL-ML 12 16.5 15 22 7194 30,303


2 SM-ML 13 19.5 15 27 12,963 38,889
3 SM-ML 14 19.5 20 31 19,084 43,860

123
20318 Neural Computing and Applications (2022) 34:20313–20325

Fig. 3 The modelling details of


the retaining system in the
software

approach, the configuration of the retaining system, bracing Table 1, and the recommended values of the developer
element, and the soil layers. [58].
The constitutive behaviour of the soil elements in the Two translational degrees of freedom are assigned to the
present study is defined by the use of the Pres- soil elements, which are fixed in a way that no relative
sureDependMultiYield02, and the PressureInde- movement is allowed in vertical translation along the lower
pendMultiYield options of the OpenSees for the sandy and boundary as well as the lateral translation on the right- and
clayey soils, respectively. The required parameters of these left-hand boundaries. Other soil elements are set free.
models are selected based on both measured parameters of

123
Neural Computing and Applications (2022) 34:20313–20325 20319

In order to set the steel piles, the central corresponding 5 Optimal design of secant pile retaining
soil elements are removed, and a hollow section tube is systems
replaced instead. The definition of the pile element in the
software is done by the use of the element dis- Basically, an optimization problem is to find the best
pBeamColumn option with 10 Gauss–Lobatto integration solution among all feasible solutions. This can be done by
points. The fibre sections are characterized by the inelastic minimizing an objective function f(x), which is subjected to
behaviour of steel S355 with an elastic modulus of 205 some constraints (i.e. inequality constraints (g(x)), and
GPA, yield strength of 355 MPa, and ultimate strength of equality constraints (h(x))) as follows:
520 MPa. Figure 3 illustrates the modelling details of the
gi ð xÞ  0i ¼ 1:2:. . .:nhj ð xÞ ¼ 0j ¼ 1:2:. . .:pLk  X k  U k k
steel pile in OpenSees software. Two translational and one
¼ 1:2:. . .:m
rotational degrees of freedom are assigned on the pile
elements, and unless the vertical translation of the bottom ð3Þ
side, all the degrees of freedom are set free. where L and U are boundary constraints.
The bracing systems (struts) are modelled in the soft- It is clear that the selection of an objective function, f(x),
ware in the X4 level to resist the applied compression force is considered as an important step in every optimization
by the excavation sides. A uni-length truss element char- problem, which directly affects the obtained results. The
acterized by the uniaxialMaterial ElasticPP is defined for objective function in these types of problems is defined by
this purpose. The S235 steel material with an elastic the cost or the weight of the system [61]. Herein, the total
modulus of 205 GPA, yield strength of 235 MPa, and cost of the structural system is defined as an objective
ultimate strength of 370 MPa is used for the strut elements. function, which is computed by Eq. (4).
The cross section of the struts is considered as a box. A
f ð xÞ ¼ C p W p þ C s W s ð4Þ
small gap in the force–deformation relationship of the
bracing element is considered to account the pile lateral where W p is the weight of pile material, W s is the weight of
deformation until that level. the strut material, and C p , and Cs are their corresponding
In order to capture the pile–soil interaction in the unit costs, respectively.
numerical modelling, the ContactMaterial2D behaviour is The constraints which control the design process can be
assigned on the BeamContact2D element. The connection classified into two main categories of behavioural and
approach of the BeamContact2D element between the pile geometric constraints. The behavioural constraints such as
and soil nodes is depicted in Fig. 3. This also allows a shear and moment capacity of structural piles, their over-
kinematic condition between the pile and the adjacent soil turning, sliding and bending capacities must satisfy the
in order to share their identical horizontal and vertical code limitation [62]. In contrast, the geometry constraints
displacements [59, 60]. The hydrostatic pressure below the are boundary constraints defining lower and upper limits of
water table is applied with triangular distribution. The each design variables. The used geometry constraints in
adjacent surface load is assumed to be 50 kN/m. this paper are listed in Table 2. The pseudo-code for design
After creating the finite element model, the static gravity of secant pile retaining systems for the considered algo-
loads are applied to establish the initial stress state in the rithms of this paper is given in Figs. 4, 5, 6, 7.
soil. Then, the excavation process is simulated by incre-
mentally removing solid layers from one side of the
retaining system. The remained is analysed for a sufficient
number of steps in a way that equilibrium is reached before
the next layer excavation. The excavation process contin- Table 2 Geometric constraints indicating lower and upper bounds
ues until the depth of - 22 m. Design variables Unit Lower bound Upper bound
For the optimization purpose, an integration of the
X1 m 0.5 2.0
OpenSees software with the MATLAB platform is done to
join the modelling space with the mentioned optimization X2 m 0.005 0.05
algorithms. The number of iterations for each run is con- X3 m 0.4 4.0
sidered to be 400, and it is assumed as a termination cri- X4 m 5.0 20.0
terion. During each iteration, the design variables are X5 m 1.0 10.0
imposed to the OpenSees software, and the outputs are sent X6 m 0.2 1.0
back to the MATLAB platform for the evaluation phase. X7 m 0.005 0.05
The optimization process is done 50 times for each X8 m 2.0 10.0
algorithm.

123
20320 Neural Computing and Applications (2022) 34:20313–20325

Fig. 4 GA pseudo-code for


design of secant pile retaining
systems

Fig. 5 PSO pseudo-code for


design of secant pile retaining
systems

6 Results and discussion According to the results of Table 3, the obtained results of
PSO, BA and GA are approximately the same and corre-
The optimal design parameters of the secant pile walls of spond to best answers, while the application of the BBO
the studied excavation pit based on the design variable algorithm leads to higher cost. However, the results of BA
boundary limitation of Table 2 are evaluated and presented during the whole analyses have lower standard deviation
in this section. For all considered optimization techniques, among the other utilized optimization algorithms which
the best, mean, and standard deviation (SD) of the objec- represents a steady performance with higher reliability
tive function are calculated and presented in Table 3. (lower variation). In contrary, the results of the PSO
algorithm correspond to higher variation comparing to

123
Neural Computing and Applications (2022) 34:20313–20325 20321

Fig. 6 BA pseudo-code for


design of secant pile retaining
systems

Table 3 Design cost values for the considered project


Optimization algorithm Cost ($)
Best Mean SD

GA 152,744.34 153,664.34 642.90


PSO 136,262.82 138,300.10 1034.25
BA 144,572.10 144,986.56 203.35
BBO 189,829.00 191,824.12 956.01

Fig. 7 BBO pseudo-code for design of secant pile retaining systems

other optimization techniques, which is not preferred in


these types of design techniques. For better comparison,
Fig. 8 The convergence rate plot for the optimal design of retaining
the results are normalized relative to the result of BA,
piles
which represents acceptable solution with lower standard
deviation. As a result, relative objective values of the GA corresponding result of PSO algorithm is 1% lower than
and BBO algorithms are 6%, 47% higher, and the the BA objective value.

123
20322 Neural Computing and Applications (2022) 34:20313–20325

Figure 8 demonstrates the convergence rates of the best depth of the piles is for the BA, while it is the minimum,
solutions obtained by the considered metaheuristic algo- considering the PSO technique.
rithms. In this figure, the vertical axis represents the total
structural cost in dollar and the horizontal axis is the iter-
ation number. As it is obvious, the PSO, GA, and BA 7 Conclusion
techniques have presented great performance in both eco-
nomical point of view and achieving rapid convergence. As The efficiency of the evolutionary optimization techniques
shown in Fig. 8, the convergence reaches in less than 200 on the optimal design of pile wall retaining structures is
iteration, and it is minimum and maximum when the GA evaluated in this article. For this purpose, the on-going
and the PSO algorithms are selected, respectively. Table 4 Tabriz metro station project is selected as the case study.
presents the final outputs of the utilized optimization The precise modelling of the pile wall retaining system as
algorithms in this paper. well as its surrounding soil layers is performed by intro-
Figure 9 investigates the pile lateral deflection during ducing a detailed finite element model, where different
the 22-stage excavations for all the utilized techniques in excavation steps are defined in order to catch its effects on
their best solutions (as given in Table 4). It is obtained that the formation of the soil shear stress. This will help to
the pile lateral deformation is increased as the excavation reduce the uncertainties of the modelling phase and gives
proceeds. During the optimization process, the maximum the possibility of monitoring the system behaviour in all the
permitted deflection of the piles is selected to be 10 cm excavation process. The GA, PSO, BA, and BBO algo-
with 20% tolerance. Comparing the results of all opti- rithms are selected in order to optimize the eight design
mization techniques, it can be concluded that the optimal variables of a considered retaining system. The objective
placement of the bracing system (struts) is achieved to be function is defined as the total structural cost, which should
in the middle-third of the excavation depth, which made be minimized in the design space of the variables. The
the maximum pile deflection to be occurred at the top of optimization process is done by integration of the Open-
the pile system. It should be noted that the utilization of a Sees software together with MATLAB platform. Although
layer of struts is sufficient for the maximum allowed pile the obtained results are limited to the Tabriz metro station
deflection. project; however, the modelling and the design procedures
The soil shear stress contours which are directly exerted can be extended to other similar projects. The main find-
to the retaining piles are depicted in Fig. 10. This fig- ings of the results can be summarized as follows:
ure demonstrates the soil shear distribution during its
• The PSO, BA and GA have been demonstrated
height and width at the final stage of the excavation.
approximately the same finesses which correspond to
According to the figure, the parts with dark red colour
best answers, while the results of the BBO algorithm
represent active performance of soil, while the rest of the
lead to higher cost.
colour ranges show soil passive performance. It is clear that
• Normalized objective values of the utilized algorithms
the maximum soil shear stress is taken place at the strut
relative to the BA’s are 6, 47% higher considering the
location, where the soil movement is hindered by the
GA, and BBO algorithms, while it is 1% lower when
bracing system.
the PSO is selected.
The pattern of soil shear stress obtained from all con-
• Considering the standard deviation of the obtained
sidered optimization techniques is approximately the same,
results, the BA leads the best answer with lower
while the values are various. It is also observed from the
variation among the other utilized optimization algo-
figure that the obtained optimal design based on the PSO
rithms, which is mostly preferred in these types of
algorithm leads to lower maximum soil shear stress, while
problems.
the contours corresponding to the GA and BBO design
• The PSO, GA, and BA techniques have presented great
variables have higher maximum shear stress values. In
performance in achieving rapid convergence.
addition, it can be also obtained that the deepest buried

Table 4 Final optimal design


Optimization algorithms X1 (m) X2 (m) X3 (m) X4 (m) X5 (m) X6 (m) X7 (m) X8 (m)
parameters of the considered
project GA 0.74 0.01 2.98 15 3 0.72 0.014 6.68
PSO 0.70 0.01 2.99 16 3 0.99 0.010 5.63
BA 0.79 0.008 2.89 18 6 0.87 0.012 6.95
BBO 1.49 0.008 3.38 13 3 0.71 0.015 5.19

123
Neural Computing and Applications (2022) 34:20313–20325 20323

Fig. 9 The retaining system lateral deflection in different excavation stages for: a GA, b PSO, c BA, and d BBO design techniques

Fig. 10 The soil shear stress in


the final excavation stage for:
a GA, b PSO, c BA, and d BBO
design techniques

• The lateral pile deflection obtained by all optimization • The optimal placement of the struts is obtained to be in
techniques is increased as the excavation depth the middle-third of the excavation depth, considering all
increases. The maximum deflection is occurred at the the applied optimization algorithms. Therefore, soil
top of the piles. shear stress pattern is approximately the same in all of

123
20324 Neural Computing and Applications (2022) 34:20313–20325

the design techniques, and the maximum soil shear multiphase drive systems. IEEE Trans Power Electr
stress is occurred at the strut location. 37(8):8767–8772
18. Asteris PG, Lourenço PB, Roussis PC, Adami CE, Armaghani
• The deepest buried depth of the piles is for the BA DJ, Cavaleri L, Pilakoutas K (2022) Revealing the nature of
optimization technique, while it is the minimum, metakaolin-based concrete materials using artificial intelligence
considering the PSO technique. techniques. Constr Build Mater 322:126500
19. Zhao C, Zhu Y, Du Y, Liao F, Chan CY (2022) A novel direct
trajectory planning approach based on generative adversarial
networks and rapidly-exploring random tree. IEEE Trans Intell
Transp Syst. https://doi.org/10.1109/TITS.2022.3164391
Declarations 20. Li B, Feng Y, Xiong Z, Yang W, Liu G (2021) Research on AI
security enhanced encryption algorithm of autonomous IoT sys-
Conflict of interest The authors declare that there is no conflict of tems. Inf Sci 575:379–398
interest. 21. Xiong Z, Mo F, Zhao X, Xu F, Zhang X, Wu Y (2022) Dynamic
texture classification based on 3D ICA-Learned filters and Fisher
vector encoding in big data environment. J Signal Proc Syst.
https://doi.org/10.1007/s11265-021-01737-0
References 22. Vakitbilir N, Hilal A, Direkoğlu C (2022) Hybrid deep learning
models for multivariate forecasting of global horizontal irradia-
1. Helwany S (2007) Applied soil mechanics with ABAQUS tion. Neural Comput Appl 34:8005–8026. https://doi.org/10.
applications. Wiley 1007/s00521-022-06907-0
2. Muthomi MA (2013) A Study of deep excavations and excava- 23. Zhang L, Zheng H, Cai G, Zhang Z, Wang X, Koh LH (2022)
tion support systems in soft soils. University of Nairobi Power-frequency oscillation suppression algorithm for AC
3. Dismuke TD (1991) Retaining structures and excavations. microgrid with multiple virtual synchronous generators based on
Foundation Engineering Handbook. Springer, pp 44–510 fuzzy inference system. IET Renew Power Gener
4. Chakraborty M, Krishna AM, Chakraborty A (2011) Reliability 16(8):1589–1601
based performance evaluation of earth retaining structures. In 24. Dong J, Cong Y, Sun G, Fang Z, Ding Z (2021) Where and how
Geo-Risk: Risk Assessment and Management 762–769. to transfer: knowledge aggregation-induced transferability per-
5. Clayton CR, Woods RI, Bond AJ, Milititsky J (2014) Earth ception for unsupervised domain adaptation. IEEE Trans Pattern
pressure and earth-retaining structures. CRC Press Anal Mach Intell. https://doi.org/10.1109/TPAMI.2021.3128560
6. Lim A, Ou CY, Hsieh PG (2020) A novel strut-free retaining wall 25. Liu F, Zhang G, Lu J (2020) Heterogeneous domain adaptation:
system for deep excavation in soft clay: numerical study. Acta An unsupervised approach. IEEE Trans Neural Netw Learn Syst
Geotech 15(6):1557–1576 31(12):5588–5602
7. Santhoshkumar G, Ghosh P (2020) Seismic stability analysis of a 26. Sun G, Cong Y, Wang Q, Zhong B, Fu Y (2020) Representative
hunchbacked retaining wall under passive state using method of task self-selection for flexible clustered lifelong learning. IEEE
stress characteristics. Acta Geotech 15(10):2969–2982 Transactions on Neural Networks and Learning Systems.
8. Fan C, Liu H, Cao J, Ling HI (2020) Responses of reinforced soil 27. Wang S, Guo H, Zhang S, Barton D, Brooks P (2022) Analysis
retaining walls subjected to horizontal and vertical seismic and prediction of double-carriage train wheel wear based on
loadings. Soil Dyn Earthq Eng 129:105969 SIMPACK and neural networks. Adv Mech Eng. https://doi.org/
9. Feng T, Liu L, Tong T, Zhou M (2017) Numerical study on 10.1177/16878132221078491
lateral wall displacement of deep excavation supported by IPS 28. Gong X, Wang L, Mou Y, Wang H, Wei X, Zheng W, Yin L
earth retention system. Underground Space 2(4):259–271 (2022) Improved four-CHANNEL PBTDPA control strategy
10. Vahedifard F, Leshchinsky BA, Mortezaei K, Lu N (2015) Active using force feedback bilateral teleoperation system. Int J Control
earth pressures for unsaturated retaining structures. J Geotech Autom Syst 20(3):1002–1017
Geoenviron Eng 141(11):04015048 29. Hosseini S, Poormirzaee R, Hajihassani M, Kalatehjari R (2022)
11. Peck RB (1969) Deep excavations and tunnelling in soft groun- An ANN-fuzzy cognitive map-based Z-number theory to predict
d.In: Proc. 7th ICSMFE 225–290 flyrock induced by blasting in Open-Pit mines. Rock Mech Rock
12. Poulos HG, Chen LT (1997) Pile response due to excavation- Eng 55:4373–4390. https://doi.org/10.1007/s00603-022-02866-z
induced lateral soil movement. J Geotechn Geoenviron Eng 30. Namazi E, Hajihassani M, Gilani SO, Zolfegharifar SY (2022)
123(2):94–99 Risk assessment of building damage induced by tunnelling
13. Gao GY, Gao M, Yang CB, Yu ZS (2010) Influence of deep through a gene expression programming model. Geotech Geol
excavation on deformation of operating metro tunnels and Eng, 1–14
countermeasures. Chin J Geotech Eng 32(3):453–459 31. Hasanipanah M, Monjezi M, Shahnazar A, Armaghani DJ,
14. Bueno BS, Benjamim CVS, Zornberg JG (2005) Field perfor- Farazmand A (2015) Feasibility of indirect determination of blast
mance of a full-scale retaining wall reinforced with Nonwoven induced ground vibration based on support vector machine.
geotextiles. In Slopes and retaining structures under seismic and Measurement 75:289–297
static conditions 1–9 32. Ceranic B, Fryer C, Baines RW (2001) An application of simu-
15. Xu P, Hatami K (2019) Sliding stability and lateral displacement late annealing to the optimum design of reinforced concrete
analysis of reinforced soil retaining walls. Geotext Geomembr retaining structures. Comput Struct 79(17):1569–1581
47(4):483–492 33. Yepes V, Alcala A, Perea C, Gonzalez-Vidosa F (2008) A
16. Hajihassani M, Armaghani DJ, Kalatehjari R (2018) Applications parametric study of optimum earth-retaining walls by simulated
of particle swarm optimization in geotechnical engineering: a annealing. Eng Struct 30:821–830
comprehensive review. Geotech Geol Eng 36(2):705–722 34. Ahmadi-Nedushan B, Varaee H (2009) Optimal design of rein-
17. Liu Z, Fang L, Jiang D, Qu R (2022) A machine-learning based forced concrete retaining walls using a swarm intelligence tech-
fault diagnosis method with adaptive secondary sampling for nique. In: Proceedings of the 1st international conference on soft

123
Neural Computing and Applications (2022) 34:20313–20325 20325

computing technology in civil, structural and environmental 48. Koopialipoor M, Murlidhar BR, Hedayat A, Armaghani DJ,
engineering. Stirlingshire (UK): Civil-Comp Press Gordan B, Mohamad ET (2020) The use of new intelligent
35. Kaveh A, Abadi ASM (2010) Harmony search based algorithm techniques in designing retaining walls. Eng Comput
for the optimum cost design of reinforced concrete cantilever 36(1):283–294
retaining walls. Int J Civil Eng 9(1):1–8 49. Kaveh A, Hamedani KB, Bakhshpoori T (2020) Optimal design
36. Khajehzadeh M, Taha MR, El-Shafie A, Eslami M (2010) Eco- of reinforced concrete cantilever retaining walls utilizing eleven
nomic design of retaining wall using particle swarm optimization meta-heuristic algorithms: a comparative study. Period Polytech
with passive congregation. Aust J Basic Appl Sci Civil Eng 64(1):156–168
4(11):5500–5507 50. Dahake AG, Saraf AS (2020) Optimal design of cantilever
37. Khajehzadeh M, Taha MR, El-Shafie A, Eslami M (2011) retaining wall using relief shelve: a review. Int J Struct Eng Anal
Modified particle swarm optimization for optimum design of 6(1):29–39
spread footing and retaining wall. J Zhejiang Univ Sci A 51. Garcı́a J, Yepes V, Martı́ JV (2020) A hybrid k-means cuckoo
12(6):415–427 search algorithm applied to the counterfort retaining walls
38. Ghazavi M, Bonab SB (2011) Learning from ant society in problem. Mathematics 8(4):555
optimizing concrete retaining walls. J Technol Educ 52. Goldberg DE (1989) Genetic algorithms in search, Optimization
5(3):205–212 and machine learning. MA: Addison-Wesley, Reading
39. Camp CV, Akin C (2012) Design of retaining walls using big 53. Kennedy J, Eberhart R (1995) Particle swarm optimization. In:
bang-big crunch optimization optimum design of cantilever Proceeding of the IEEE on international conference on neural
retaining walls. J Struct Eng 138:438–448 networks. Perth, Australia 1942–8
40. Kaveh A, Behnam AF (2013) Charged system search algorithm 54. Paian Purba SR, Amarilies HS, Rachmawati NL, Perwira Redi
for the optimum cost design of reinforced concrete cantilever AAN (2021) Implementation of particle swarm optimization
retaining walls. Arab J Sci Eng 38:563–570 algorithm in cross-docking distribution problem. Acta Inform
41. Sheikholeslami R, Gholipour Khalili B, Zahrai SM (2014) Malays 5(1):16–20
Optimum cost design of reinforced concrete retaining walls using 55. Yan Z (2018) Artificial Bee Colony Constrained Optimization
hybrid firefly algorithm. Int J Eng Technol 6(6):465 Algorithm with Hybrid Discrete Variables and Its Application.
42. Gandomi AH, Kashani AR, Roke DA, Mousavi M (2015) Opti- Acta Electr Malays 2(1):18–20
mization of retaining wall design using recent swarm intelligence 56. DT Pham A Ghanbarzadeh E Koc S Otri S Rahim M Zaidi (2005)
techniques. Eng Struct 103:72–84 The Bees Algorithm. Technical Note, Manufacturing Engineering
43. Aydogdu I (2017) Cost optimization of reinforced concrete Centre, Cardiff University, UK
cantilever retaining walls under seismic loading using a bio- 57. Simon D (2008) Biogeography-based optimization. IEEE Trans
geography-based optimization algorithm with Levy flights. Eng Evol Comput 12(6):702–713
Optim 49(3):381–400 58. Yang Z, Lu J, Elgamal A (2008) OpenSees soil models and solid-
44. Kaveh A, Hamedani KB, Zaerreza A (2020) A set theoretical fluid fully coupled elements user’s manual
shuffled shepherd optimization algorithm for optimal design of 59. Randolph MF, Wroth CP (1981) Application of the failure state
cantilever retaining wall structures. Eng Comput 37:3265–3282. in undrained simple shear to the shaft capacity of driven piles.
https://doi.org/10.1007/s00366-020-00999-9 Geotechnique 31(1):143–157
45. Konstandakopoulou F, Tsimirika M, Pnevmatikos N, Hatzigeor- 60. Loukidis D, Salgado R (2008) Analysis of the shaft resistance of
giou GD (2020) Optimization of reinforced concrete retaining non-displacement piles in sand. Géotechnique 58(4):283–296
walls designed according to european provisions. Infrastructures 61. Saribas A, Erbatur F (1996) Optimization and sensitivity of
5(6):46 retaining structures. J Geotech Eng 122(8):649–656
46. Gordan B, Koopialipoor M, Clementking A, Tootoonchi H, 62. Iranian National Building Codes Compilation Office (2014) Ira-
Mohamad ET (2019) Estimating and optimizing safety factors of nian National Building Code, Part 9: Reinforced Concrete
retaining wall through neural network and bee colony techniques. Buildings Design, Ministry of Housing and Urban Development
Eng Comput 35(3):945–954 (MHUD)
47. Mergos PE, Mantoglou F (2020) Optimum design of reinforced
concrete retaining walls with the flower pollination algorithm. Publisher’s Note Springer Nature remains neutral with regard to
Struct Multidiscip Optim 61(2):575–585 jurisdictional claims in published maps and institutional affiliations.

123

You might also like