Field Performance of a Genetic Algorithm in the Settlement Prediction of a
Thick Soft Clay Deposit in the Southern Part of the Korean Peninsula
Hyun Il Park, Kyu-Sun Kim, Ha Young Kim
PII:
DOI:
Reference:

S0013-7952(15)30021-1
doi: 10.1016/j.enggeo.2015.07.012
ENGEO 4104

To appear in:

Engineering Geology

Received date:
Revised date:
Accepted date:

27 August 2014
15 July 2015
16 July 2015

Please cite this article as: Park, Hyun Il, Kim, Kyu-Sun, Kim, Ha Young, Field Performance of a Genetic Algorithm in the Settlement Prediction of a Thick Soft Clay
Deposit in the Southern Part of the Korean Peninsula, Engineering Geology (2015), doi:
10.1016/j.enggeo.2015.07.012

This is a PDF le of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its nal form. Please note that during the production process
errors may be discovered which could aect the content, and all legal disclaimers that
apply to the journal pertain.

ACCEPTED MANUSCRIPT
Field Performance of a Genetic Algorithm in the Settlement Prediction of a Thick Soft Clay

PT

Deposit in the Southern Part of the Korean Peninsula

RI

Hyun Il Park 1, Kyu-Sun Kim 2, and Ha Young Kim 3

SC

Abstract

NU

Long-term settlement data of a thick soft clay deposit improved by vertical drains frequently
shows different field settlement behavior from laboratory-driven predictions by conventional

MA

theories due to the variability and uncertainty of the soil properties, modeling simplifications,
and types of prediction methods. This paper presents the field application of a back-analysis

TE

method based on a genetic algorithm (GA) to evaluate the performance of a new settlement
prediction method compared with conventional graphical settlement prediction methods, such as

AC
CE
P

the hyperbolic method and the Asaoka method. The GA back-analysis method shows better
flexibility in modifying surcharging plans and adaptability to multi-layered thick soft soil
deposits at the early stages of post-construction settlements. Thus, this new settlement prediction
method enables geotechnical engineers to subsequently modify the heights of surcharge fills
subsequently in accordance with field settlement data monitored in the interim for rapid and costeffective construction. The comparative results show that the GA back-analysis method is
capable of superior field performance in settlement predictions compared with two conventional

Principal Research Engineer, Construction Technology Division, Samsung C&T Corporation, Seoul 135-935
Korea, Email: gomdori7@empas.com
2
Principal Research Engineer, Construction Technology Division, Samsung C&T Corporation, Seoul 135-935
Korea. E-mail: kyusun@empas.com (Corresponding author)
3
General Manager, Technical Advisory Team, Samsung C&T Corporation, Seoul 137-956 Korea. E-mail:
hy05.kim@samsung.com

ACCEPTED MANUSCRIPT
graphical methods, within a margin of less than 200 mm in a thick soft clay deposit with multiple
layers under complex loading conditions.

RI

PT

Highlights
We assessed the field performance of the GA method in settlement prediction.
The GA method shows a higher performance in a multi-layered thick soft soil deposit.
The GA method predicts the final settlement at the early stage of construction.
Keywords
Back analysis; Consolidation; Genetic algorithm; Settlement prediction; Soft clay; Vertical drain

SC

1. Introduction

NU

Over the past two decades, several construction projects in Korea have been built on thick soft
soil deposits to create new national infrastructure, such as container ports and a new international

MA

airport, which were built on reclaimed areas. As a cost-effective soil improvement method,
embankment preloading associated with vertical drains is commonly used to accelerate the

consolidation process and predict the consolidation settlements of foundation soils due to

TE

surcharge loading. Settlement prediction in a large-construction project is an important issue for

AC
CE
P

rapid and cost-effective construction; however, it is frequently reported that geotechnical

engineers fail to obtain accurate predictions of the long-term field settlements of soft soils even if
a robust calculation with advanced geotechnical technology was implemented during the design
stage. Significant errors in settlement predictions during ground improvement work have been
reported in many construction sites in Korea. Recent construction records in the southern part of
Korea indicate that the observed settlements are is generally 1.5 to 3 times greater than the
predicted settlements predicted in the design stage. It is thought expected that these errors in
settlement prediction errors are mainly caused by the variability and uncertainty of the soil
properties (Kim, 2008; Chung et al., 2009; Park et al., 2009; Chung et al., 2012).
Although the accuracy of numerical schemes and testing techniques has been significantly
improved, there are still errors between the predicted and observed magnitudes of settlement
during a prediction and during an observation. Possible reasons for these errors are the following
2

ACCEPTED MANUSCRIPT
as follows: (1) sampling disturbances and effective stress condition during sampling, delivery,
and trimming procedures; (2) a simplified analysis model, loading condition, boundary condition,

PT

and initial condition used during the design stage; (3) measurement errors caused by operators

RI

and systems; and (4) variability and uncertainty of the soil properties (Bjerrum, 1967;
Jamiolkowski et al., 1985; Terzaghi et al., 1996). Graphical methods using observed settlement

SC

data are commonly used employed as practical tools for interpreting field settlement data to

NU

reduce the variability and uncertainty of the consolidation parameters governing the magnitude
of the consolidation settlement (Asaoka, 1978; Tan and Chew, 1996).

MA

Soil improvement performance levels are usually evaluated by analyzing analysis of the
monitored settlement data for comparison with laboratory-driven settlements determined during

TE

the design stage. The magnitude of the additional surcharge load and the surcharge period can be
finally determined based on the interpretation of field settlement data. The consolidation

AC
CE
P

parameters in a thick soft clay deposit with vertical drains are the main points of interest in
predicting long-term consolidation settlements. The consolidation settlement of a vertical drain
installed in a soft soil layer is calculated based on Barrons equation. Several observational
methods based on settlement records are available to predict future settlement behavior,
specifically the hyperbolic method and the Asaoka method. By extrapolating extrapolation from
observed settlement data, many uncertainties regarding the variability of the soil and as well as
the magnitude and the distribution of loads can be overcome (Tan et al., 1991; Asaoka, 1978).
These graphical methods simply predict the final settlements by means of curve fitting from
observed settlement data. Therefore, observed settlement data after removal of surcharge loads
are used to predict the final settlements related to the degree of consolidation. However, these
methods are not suitable for use when predicting field settlements during unloading and

ACCEPTED MANUSCRIPT
reloading after the removal of the surcharge load and when adjusting the height and period of the
surcharging embankment (Park et al., 2009).

PT

The BFGS (Broyden-Fletcher-Goldfarb-Shanno) method, the quasi-Newton method, and the

RI

conjugate gradient method are conventional optimization methods for consolidation-related

back-analysis problems (Arai et al., 1983; Sakurai and Takeuchi, 1983; Shoji et al., 1990; Kim

SC

and Lee, 1997; Park et al., 2009). These methods have been are used for consolidation problems

NU

with a small numbers of consolidation parameters in one or two clay layers. However, the
number of consolidation parameters in a multi-layered problem is typically greater than that in a

MA

single layer problem. This multi-dimensional optimization problem complicates the process of
estimating consolidation parameters when attempting to predict the consolidation settlement of

TE

foundation soils accurately. Several optimization schemes generate either local or global
solutions depending on the initial values of the variables used. Furthermore, as the number of the

AC
CE
P

variables for the optimization process increases, the chance of a solution converging locally
rather than globally increases (Renders and Flasse, 1996; Leung and Wang, 2001). Therefore, in
a multi-layered soft clay deposit, it is necessary to use a robust technique that guarantees
convergence to a global solution. A back-analysis method based on a genetic algorithm (GA) can
be used as a parallel and global search tool that emulates natural genetic operators. GAs
generally show better performance when searching for a solution than conventional optimization
algorithms because GAs, which make use of an entire set of solutions spread throughout the
solution space, are less affected by local optima (Holland, 1975; Goldberg, 1989; Keane, 1995).
Park et al. (2009) showed that the GA back-analysis method has the advantage of robustly
searching for a global solution while avoiding local solutions compared to with conventional

ACCEPTED MANUSCRIPT
optimization schemes in a multi-dimensional consolidation problem with three consolidation
layers.

PT

In this study, the GA back-analysis method was implemented to improve the field

RI

performance of settlement predictions. The accuracy and capability of the settlement prediction
methods were investigated to evaluate the performance of settlement predictions settlement

SC

prediction performance using field settlement data from a large construction site in Busan on

NU

thick soft clay layers consisting of multiple-clay layers with a complex geological history.

MA

2. Site Description
2.1 Soil Characterization

TE

The studied site is located near in the southern coastal area of the Korean peninsula. The site
was mainly formed by an accumulation of a deltaic deposit near the Nakdong River. The thick

AC
CE
P

deltaic deposit area was formed as a valley caused by the vertical movement of the Yangsan fault
during the Cenozoic era. Various types of geo-materials have been deposited in the valley via the
river stream flow (Kim, 2008). A representative sectional profile near the studied site is shown in
Figure 1. The sand Sand and gravel materials were deposited in the valley first. , and then, Then,
the soft soil deposit was followed by the deposition of different geo-materials affected by marine
transgression and regression during a the area's geological history. The sedimentary deposit can
be categorized into four layers: upper clay, silty sand, lower clay, and bottom sand and gravel
layers. The bottom sand and gravel layer is thick enough to hold fresh water in the an aquifer.
The silty sand layer in the middle is sandwiched by between the upper and lower clay layers.
Soft soil layers with different engineering characteristics have been deposited by affecting by the
flow of the river water, as well as the marine transgression and regression. The soft soil layers

ACCEPTED MANUSCRIPT
can be categorized into two distinct silty clay layers as the upper and lower clay layers. The basic
characteristic of the upper clay layer basically shows a normally consolidated state with a soft to

PT

medium consistency; however, the lower clay layer shows stiff to very stiff characteristics. The

RI

total thickness of the whole upper and lower clay layers ranges from approximately 50 m to 80 m
(Chung et al., 2002; 2012; Kim, 2008).

SC

The silty clay soils in the studied site are generally classified as CL and CH soils in

NU

accordance with the Unified Soil Classification System (USCS). Figure 2 shows the profiles of
liquid and plastic limits, water content, initial void ratio, compression index, and

MA

overconsolidation ratio of the upper and lower clay layers. There is a distinct separation around
the expected border of the upper and lower clay layers in the soil profiles. The values tend to

TE

increase with depths above 25 m; however, they tend to reversely decrease below depths of 25 m.
When considering this distinct difference between two groups of data in the property profiles, the

AC
CE
P

clay layer can be divided into two representative layers as the upper clay layer and lower clay
layer. The laboratory-driven design values of the compression index (Cc) are 0.9 for the upper
clay layer and 0.6 for the lower clay layer. The laboratory-driven horizontal coefficient of
consolidation (ch) is 8.5 10-3 m2/day for both the upper and lower clay layers. The
overconsolidation ratio of the upper clay layer ranges from 1.0 to 3.0 with depth, indicating
normally consolidated to lightly overconsolidated conditions, respectively. Generally, the
overconsolidated conditions of the upper clay layer are assumed to be normally consolidated due
to the conservative design in estimating the magnitude of consolidation settlement.

2.2 Construction Plan of Studied Site

ACCEPTED MANUSCRIPT
The studied site is located near the city of Busan, which is the second largest city in South
Korea. Busan has the largest port facilities in Korea, including the recently constructed Busan

PT

New Port, which is placed was built on the thick deltaic deposit near the Nakdong River. The

RI

studied site included construction work regarding three container terminals in Busan New Port.
To accelerate the rate of ground settlement of the thick soft soil deposit, prefabricated vertical

SC

drains with embankment preloading were applied to the site. The thickness of the soft soil layers

NU

was variable, ranging from 5 m to 60 m with different depositional environments. The

arrangement distance between the vertical drains was 1.2 m, and they were arranged in a square

MA

pattern (S = 1.2 m). , and the The design load ranged from 35 kPa to 80 kPa. The design criteria
to determine the removal time of the surcharge load were controlled by according to the degree

TE

of consolidation.

Embankment preloading is a popular soil improvement method that accelerates consolidation

AC
CE
P

settlement and involves placing a surcharge fill on top of the permanent embankment. In addition,
preloading has the advantage of reducing long-term settlement if the surcharge load is left in
place long enough to achieve a significant change in the soil properties. However, additional
surcharge loads, which were not have not been considered in the design stage, are frequently
adapted to reduce the post construction settlement and to accelerate the consolidation process,
regardless of the original design plan. Therefore, geotechnical engineers may be more likely to
have complicated problems in settlement predictions caused by a modified surcharge plan during
construction. To monitor field settlement data, square-shaped steel settlement plates were
installed and used to monitor the variations of in settlements on the original ground surface at the
specified 13 locations. Ground settlements induced by embankment preloading were was
monitored by taking optical leveling measurements from a riser pipe attached to the settlement

ACCEPTED MANUSCRIPT
plates. Two graphical settlement prediction methods (i.e., hyperbolic method, and Asaoka
method) and the GA back-analysis method were implemented to compare the results of

PT

settlement predictions results. The field settlements were was monitored for about 540 days for

RI

this comparative performance evaluation in of settlement predictions.

SC

3. Observational Methods in Predicting Ground Settlement

NU

In preloading embankments with vertical drains on soft soil deposits, differences between
observed and predicted settlements frequently are induced from by the variability and uncertainty

MA

of the soil properties. To supplement address the these differences, the degree of consolidation
should be determined by the interpretation of the observed settlement data using graphical

TE

methods or a back-analysis method. In this study, observed settlement data obtained from in a
construction site were used to predict final settlements by using the GA back-analysis method

AC
CE
P

(Park et al., 2009), the conventional hyperbolic method (Tan et al., 1991), and the Asaoka
method (Asaoka, 1978). Semi-empirical graphical methods (i.e., hyperbolic method, and
Asaokas method) are practical in for settlement predictions because they rely on observed
settlement data to predict the end of primary consolidation, which can be updated as more
observed settlement data become available using simple calculation and curve fitting. However,
the conventional methods sometimes have limitations in some aspects of the time interval of
observed settlement data and the minimum period duration of monitoring duration (Edil et al.,
1991; Tan et al., 1991). The hyperbolic method can predict final settlement using settlement data
at the start of construction; however, settlement predictions for long-term field behavior using
the early-stage settlement data sometimes mislead skew the results of settlement predictions.
Thus, Therefore, before using the hyperbolic method, it is recommended to collect settlement

ACCEPTED MANUSCRIPT
data when the data reach more than a 60% degree of consolidation before using the hyperbolic
method. The Asaoka method uses the end of construction settlement data, and settlement

PT

predictions can only be made after the end of construction; thus, hence, the Asaoka method takes

RI

a longer time to predict future settlement compared with the hyperbolic method (Tan and Chew,
1996; Chung et al., 2009). Thus, the performance of the selected observational methods should

SC

be carefully evaluated from the characteristic of the curve plotted.

NU

A back-analysis technique has been used as an alternative tool to replace conventional semiempirical methods not only to accurately predict final settlements but also to inversely evaluate

MA

back-analyzed parameters (Sakurai and Takeuchi, 1983; Gioda and Sakurai, 1987). The
differences between the results of numerical simulation to estimate soil behavior and the actual

TE

measured data are defined as an objective function; then, the back-analysis technique inversely
estimates input parameters using an optimization scheme to minimize the value of the objective

AC
CE
P

function. The consolidation parameters containing actual field settlement behavior can be
estimated through a back-analysis procedure using measured settlement data. Through this
procedure, the design quality can be evaluated by estimating residual settlements by loading and
unloading, and the optimized design is conducted determined by optimization of the height and
period for surcharge loading. Park et al. (2009) proposed the GA back-analysis method to
inversely analyze multiple design parameters for multi-layered consolidation settlement data and
then evaluated the performance of the proposed method with other optimization schemes for
model problems. In this study, the collected field settlement data collected during the
consolidation process were interpreted by graphical fitting methods and the GA back-analysis
method to evaluate the magnitudes of final settlements of thick soft soil layers caused by
embankment preloading.

ACCEPTED MANUSCRIPT

3.1 Settlement Prediction Model in Soft Soils Improved by Vertical Drains

PT

Barron (1948) proposed an analytical solution to calculate the consolidation settlement of a

soil layer where a vertical drain is installed that considers free-strain and equivalent-strain

RI

constraints based on Terzaghis one-dimensional consolidation theory. The consolidation

SC

analysis based on the equivalent strain condition assumes that the initial surcharge loading is all

NU

applied as excess pore water pressure, that the strain develops in a vertical direction, and that the
influence boundary of the vertical drainage is circular. The time factor of consolidation (Th) is

MA

determined from Eq. (1-a), and the relationship between the average degree of consolidation
(Uavg) and the time factor of consolidation is expressed in Eq. (1-b). Using these relations, the

TE

ch t
d e2

(1-a)

AC
CE
P

Th

final settlement (sf) of a soft clay deposit is calculated from Eq. (1-d) in the design stage.

8Th
U avg 1 exp

F(n )
n2
3n 2 1

ln(
n
)

n2 1
4n 2

(1-c)

'
Cc

H log 0 '

1 e0
0

(1-d)

F(n )

sf

(1-b)

where n is the spacing ratio (n = de/dw = 27.2), ch is the horizontal consolidation coefficient, Th is
the time factor, de is the equivalent diameter of the drain (de = 1.13S = 1.36 m), dw is the
drainage diameter (dw = 0.05 m), Uavg is the average consolidation ratio, F(n) is the drainage
spacing influence factor, sf is the final settlement, Cc is the compression index, e0 is the initial

10

ACCEPTED MANUSCRIPT
void ratio, H is the thickness of clay layer, r0 is the initial effective stress, 0 is the effective

PT

stress, and is the surcharge load.

RI

3.2 Hyperbolic Method

The hyperbolic method is the most popular settlement prediction method. The method

SC

assumes that the shape of the future settlement curve is hyperbolic. Since a hyperbolic

NU

relationship can generate future settlement behavior with only the initial settlement data, this
method can be a powerful tool for settlement predictions. However, geotechnical engineers need

MA

to make sure to must collect sufficient settlement data to improve the accuracy of their prediction
in the final settlement to prevent unexpected excess settlement at the end of construction (Tan et

TE

al., 1991; Tan, 1993; 1995).

The ratio of elapsed time (t-t0) and the settlement during elapsed time (st-s0) can be plotted as

AC
CE
P

shown in Fig. 3(a). , and then, Then, slope and intercept in the y-axis are determined by a
linear regression. Finally, the final settlement (sf) is calculated by Eq. (2-b):
t
t
st s0

sf s0

(2-a)

(2-b)

where st is the settlement at time t, s0 is the settlement at the end of surcharging at t0, and and
are the empirical constants of the hyperbolic equation.

3.3 Asaoka Method

Asaoka (1978) proposed a graphical method to predict the final settlement (sf) based on an
observation procedure. The Asaoka method has been broadly used to predict the magnitude and

11

ACCEPTED MANUSCRIPT
time rate of consolidation using settlement data at a constant time interval under marine clay or
dredged reclaimed soils with vertical drains (Choa et al., 1981; Bo et al., 1997; 1998; Arulrajah

PT

et al., 2004). This method can be a useful tool due to its simplified procedure; however, the

RI

accuracy of the Asaoka method strongly depends on the choice of time interval during the
graphical procedure (Edil et al., 1991).

SC

When settlement data are plotted as a settlement curve throughout the duration of surcharge

NU

loading, the settlement curve is divided into a certain time interval, and the corresponding
settlements (si) are tabulated. Settlement at a specific time (st) is plotted against the settlement

MA

from the preceding time. If the settlement data are plotted in a straight line, the final settlement
under surcharge loading can be determined by extending the straight line through the data until it

TE

intersects a 1:1 line. The intersection point indicates the final settlement (sf), as shown in Fig.
3(b):

AC
CE
P

s i 0 i s i 1

sf

(3-a)

0
1 1

(3-b)

where si is the measured settlement at time i, and 0 and 1 are the intercept and slope of the
plotted line, respectively.

3.4 Genetic Algorithm Based Back Analysis Method

Park et al. (2009) introduced an application of the GA back-analysis method in the settlement
calculation based on Barrons solution. The objective function was defined as the square of the
difference between the observed settlement and the settlement calculated by Barrons solution.
The most dominant consolidation parameters in the settlement calculation (i.e., compression

12

ACCEPTED MANUSCRIPT
index, and coefficient of consolidation) were selected for the optimization procedure using a
back calculation.

PT

The GA is a methodology using that uses biological evolutionary theory to overcome local

RI

convergence during the optimization process. A global and robust solution can be obtained by
using this optimization process based on the mechanism of natural genetics and selection

SC

(Holland, 1975; Goldberg, 1989). The GA procedure used in this study is illustrated in the

NU

flowchart of Fig. 4. Detailed The step-by-step procedures is summarized in detail are as follows:
(1) Initialize the population of candidate individuals using real number chromosomes.

MA

(2) Calculate the objective function for each individual.

(3) Calculate the fitness value of each individual by the rank-based fitness assignment.

TE

(4) Select individuals and place them in the mating pool in accordance with the rank-based
fitness assignment and stochastic universal sampling.

AC
CE
P

(5) Conduct reproduction and mutation on the current population for new individuals.
(6) Insert a number of new individuals randomly replacing old individuals in the current
population. Additionally, special care should be taken not to replace the best individual
in the population.

(7) Calculate the objective function and the fitness value for each individual.
(8) The steps 3 to 7 are called a generation, and these steps are repeated until a specified
stopping criterion is satisfied. A typical stopping criterion in a genetic algorithm is a
predefined maximum number of generations.
Through this GA procedure, the best individual in the final generation will be selected as the best
solution (Park et al., 2009).

13

ACCEPTED MANUSCRIPT
To minimize the objective function in Eq. (4), each consolidation parameter in each layer is
iteratively optimized using the GA procedure. After iterative optimization procedures are carried

PT

out to find the optimized consolidation parameters, a settlement curve can be generated by using

RI

the back-analyzed parameters. If the degree of consolidation at the time of surcharge removal
does not satisfy the design criteria, changing the applied applying load condition needs to should

SC

be considered to change for the additional surcharge load or time of surcharge removal.
Num

NU

Objective function, Obj V S(i) S (i)

(4)

i 1

MA

where Num is the number of measured data, S(i) is the i th measured settlement, and S (i) is the

i th estimated settlement estimated by Barrons solution.

TE

4. Interpretation of Genetic Algorithm Based Back Analysis

AC
CE
P

4.1 Optimization for Consolidation Parameters Using Genetic Algorithm

Barrons model requires two consolidation parameters, the compression index (Cc) and the
horizontal consolidation coefficient (ch), for a single clay layer. Double The double clay layers
(i.e., upper clay layer, and lower clay layer) shown in Fig. 2 need require four consolidation
parameters. To determine the accurate consolidation parameters of a clay deposit from the
settlement pattern, it is essential to apply an appropriate optimization technique that provides a
global solution. This is even more important if the number of the consolidation parameters
increases during the optimization process. Basic parameters for the GA back analysis are
summarized in Table 1.
Figure 5 shows the optimization procedure of the four consolidation parameters of the upper
and lower double-layered clay deposit by elapsing generation numbers. When a generation
number is greater than 15, the design variable tends to be converged. In this study, the optimized
14

ACCEPTED MANUSCRIPT
value was determined at the generation number of 30 to obtain consistent results. Figure 6 shows
the variations of consolidation parameters during the optimization procedure by back calculation

PT

with a 30-day interval. After 220 days (U 40%), the induced consolidation parameters tended

RI

to be constant regardless of measurement periods.

Figure 7 shows the comparative compares the results of the predicted settlement curve by

SC

design, measurement, as well as the back-analysis stages depending on in relation to a specified

NU

time period. The GA back-analysis method predicts predicted the settlement behavior very well
compared to with the measured settlement data. The difference between the predicted final

MA

settlements predicted by back calculation at a 35% degree of consolidation and back calculation
at a 65% degree of consolidation is was less than 200 mm. This result shows that the GA back-

TE

analysis method shows a has better prediction capability at the early stage of the postconstruction settlement even if the degree of consolidation is less than 50%.

AC
CE
P

The back-analyzed parameters at each settlement plate are summarized in Table 2. The
results show that the consolidation parameters induced from the field settlement data are were
significantly different from the laboratory-driven parameters. In the upper clay layer, the backanalyzed compression index is was generally underestimated in comparison with the laboratorydriven compression index; however, the results in the lower clay layer showed a reverse trend. In
addition, the coefficient of consolidation is was generally overestimated in at the design stage
compared with the laboratory-driven parameter. The back-analyzed compression index of the
upper clay layer ranges ranged from 0.18 to 0.58, and that of the lower clay layer ranges ranged
from 0.71 to 1.10. The compression indexes of the upper clay layer and lower clay layer at the
design stage were originally 0.9 and 0.6, respectively. However, the results of the back analysis
showed a reverse trend compared with the design values. This reverse trend includes complex

15

ACCEPTED MANUSCRIPT
interactions on the mechanism of consolidation. The mismatching trend may be mainly caused
by due to a conservative design concept and sampling disturbances. The upper clay layer is

PT

generally assumed to be normally consolidated condition for the in a conservative design

RI

allowing a sufficient amount of settlement; however, the actual condition of the layer may be in
lightly overconsolidation overconsolidated condition in the of a shallow depth as shown in the

SC

OCR profile of Fig. 2. Thus, the lightly overconsolidated condition of the upper clay layer is

NU

reflected to in the underestimated back-analyzed compression index compared with the design
value. On the other hand, the lower clay layer shows a normally consolidated condition; however,

MA

the effect of sampling disturbances of the lower clay layer during the design stage may be a the

TE

main reason for the overestimated back-analyzed value.

4.2 Field Performance of Settlement Prediction Methods

AC
CE
P

To evaluate the performance of the settlement prediction methods, final settlements predicted
prediction made by the hyperbolic method, the Asaoka method, and the GA back-analysis
method were compared with the recorded settlement data recorded in at the Busan New Port
construction site for up to 560 days. The predicted final settlements predictions at the 13
locations made by settlement prediction these methods are summarized in Table 3. The results
show that the hyperbolic method estimates estimated the greatest final settlements, followed by
the GA back-analysis method, and the Asaoka method. Tan and Chew (1996) also reported that
the hyperbolic method usually predicted the most conservative final settlement compared with
the final settlements predictions made by the other methods.
As shown in Fig. 8, the final settlements predicted by the GA back-analysis method are was
generally less than the final settlements predicted by the hyperbolic method and greater than the

16

ACCEPTED MANUSCRIPT
measured settlements. On the other hand, the Asaoka method generally underestimates
underestimated the final settlement, which may cause lead to the application of an additional

PT

surcharge fill to satisfy the design criterion. Thus, when geotechnical engineers consider

RI

determining the completion of a consolidation settlement of clay deposit (e.g., U > 90%), they
need to know the characteristics of a selected settlement prediction method in order to avoid

SC

additional construction cost. The comparative results A comparison of the performance in

NU

settlement predictions performance results shows that the GA back-analysis method most
accurately predicts provides the most effective results in predicting the final settlement. The GA

MA

back-analysis method provides the lowest amount of error variance when predicting final
settlement amounts (< 200 mm), followed by the Asaoka method (< 500 mm) and the hyperbolic

TE

method (< 600 mm).

AC
CE
P

5. Field Application of Modified Surcharging Plan

Field geotechnical engineers are required to check and modify a design surcharging plan
during construction to reflect the actual field conditions in the of settlement management. Thus,
Therefore, they should check and predict whether the field settlement will be completed with
achieving above 90% consolidation using a reliable settlement prediction method. The beauty of
the GA back-analysis method is its flexibility to easily modify the surcharging fill using the
back-analyzed consolidation parameters. For example, the original and modified surcharging
plans and the settlement curves of the plate SP-12 are shown in Fig. 9. One-hundred and ninety
days after loading, field settlement was greater than the showed excessive settlement compared
to settlement predicted in the design stage. Thus, it was expected that the settlement could not
satisfy the design criterion of 90% consolidation settlement within the given construction period.

17

ACCEPTED MANUSCRIPT
Thus, the height of the surcharge fill was recalculated using the consolidation parameters backanalyzed from the settlement measured for 190 days using the GA. and then, Then, it was

PT

determined that an additional 5 m surcharge fill would be required to satisfy the design criterion

RI

of 90% consolidation settlement within a specified construction period. The final result shows
that the GA back-analysis method can modify the surcharging plan and predict future settlement

SC

behavior using back-analyzed consolidation parameters at the early stage of construction (260

NU

days, U < 50%).

MA

6. Conclusion

A comparative evaluation of the performances of settlement prediction methods was

TE

conducted for a site on a multi-layered thick soft clay deposit in Korea. Settlement data from 13
settlement plates at the site during embankment preloading were used to evaluate the accuracy of

AC
CE
P

the settlement prediction methods. The results showed that the GA back analysis method offers
better accuracy when used to predict the final settlement even when the surcharging plan is
modified at the early stage of construction. The specific findings of this study are summarized as
follows:

1) Among the three settlement prediction methods assessed, the hyperbolic method tends to
overestimate the final settlements compared to with the other two methods; however, the
variances of the predicted settlements in the hyperbolic method are smaller than those in the
Asaoka method.
2) In the upper clay layer, the back-analyzed compression index is generally underestimated in
comparison with the laboratory-driven compression index; however, the results in the lower
clay layer show a reverse trend. In addition, the coefficient of consolidation is generally

18

ACCEPTED MANUSCRIPT
overestimated in the design stage compared with the laboratory-driven parameter. These
discrepancies may be caused by due to the conservative design concept in the lightly

PT

overconsolidated region and the effect of sampling disturbances.

RI

3) The GA back-analysis method gave achieved the most reliable and consistent settlement
predictions for most of the 13 settlement records within a margin of less than 200 mm

SC

compared with recorded settlements measurements.

NU

4) The GA back-analysis method shows the lowest amount of error variance when predicting
final amounts of settlement amounts (< 200 mm), followed by the Asaoka method (< 500

MA

mm) and the hyperbolic method (< 600 mm).

5) The GA back-analysis method can predict the final settlements with back-analyzed

surcharging plan.

TE

parameters at an early stage of construction (U < 50%) regardless of modifications to the

AC
CE
P

The overall results showed that the GA back-analysis method offers better accuracy when used
to predict the final settlement even when the surcharging plan is modified at the early stage of
construction. However, a small number of variables (Cc and ch) in each layer were implemented
to the GA back-analysis method of this study to minimize the calculation time for the practical
use of a settlement management tool at the construction site. Thus, if we need to obtain more
accurate settlement predictions, we may consider using more variables during the GA back
analysis to predict much closer curve-fitting to the actual settlement behavior under complex
loading conditions on multi-layered soft soil deposits.

Acknowledgements

19

ACCEPTED MANUSCRIPT
The authors acknowledge the support of the Construction Technology Division of Samsung
C&T Corporation. The authors appreciate the constructive comments and suggestions from the

RI

PT

anonymous reviewers.

References

SC

Arai, K., Ohta, H., Yasui, T., 1983. Simple optimization techniques for evaluating deformation

NU

moduli from field observations. Soils Found. 23(1), 107-113.

Arulrajah, A., Nikraz, H., Bo, M.W., 2004. Factors affecting field instrumentation assessment of

MA

marine clay treated with prefabricated vertical drains. Geotext. Geomembranes. 22(5), 415437.

TE

Asaoka, A., 1978. Observation procedure of settlement prediction. Soils Found. 18(4), 87-101.
Barron, R.A., 1948. Consolidation of fine-grained soils by drain wells. Trans. ASCE 113(2346),

AC
CE
P

718-754.

Bjerrum, L., 1967. Engineering geology of Norwegian normally consolidated marine clays as
related to settlements of buildings. Geotechnique 17(2), 81-118.
Bo, M.W., Arulrajah, A., Choa, V., 1997. Assessment of degree of consolidation in soil
improvement project. Proc. Intl. Conf. on Ground Improvement Techniques, Macau, pp. 71
80.
Bo, M.W., Arulrajah, A., Choa, V., Chang, M.F., 1998. Site characterization for a land
reclamation project at Changi in Singapore. Proc. 1st Intl. Conf. on Site Characterization,
Atlanta, pp. 333-338.
Choa, V., Karunaratne, G., Ramaswamy, S.D., Vijiaratnam, A., Lee, S.L., 1981. Drain
performance in Changi marine clay. Proc. 10th Intl. Conf. on Soil Mech. & Fdn. Eng.,

20

ACCEPTED MANUSCRIPT
Stockholm, Vol. 3, pp. 623-626.
Chung, S.G., Giao, P.H., Kim, G.J., Leroueil, S., 2002. Geotechnical properties of Pusan clays.

PT

Can. Geotech. J. 39(5), 1050-1060.

RI

Chung, S.G., Ryu, C.K., Min, S.C., Lee, J.M., Hong, Y.P., and Odgerel, E., 2012. Geotechnical
characterisation of Busan clay. KSCE J. Civil Eng. 16(3), 341-350.

SC

Chung, S.G., Lee, N. K., Kim, S.R., 2009. Hyperbolic method for prediction of prefabricated

NU

vertical drains performance. J. Geotech. Geoenviron. Eng. 135(10), 1519-1528.

Edil, T.B., Fox, P.J., and Lan, L.T., 1991. Observational procedure for settlement of peat. Proc.

MA

Geo-Coast 91 Conf., Yokohama, pp. 165-170.

Gioda, G., Sakurai, S., 1987. Back analysis procedures for the interpretation of field

TE

measurements in geomechanics. Int. J. Num. Anal. Met. 11(6), 555-583.

Goldberg, D.E., 1989. Genetic Algorithms in Search, Optimization, and Machine Learning.

AC
CE
P

Addision-Wesley, Boston..

Holland, J.H., 1975. Adaptation in Natural and Artificial Systems. University of Michigan Press,
Ann Arbor.

Jamiolkowski, M., Ladd, C.C., Germaine, J.T. & Lancellotta, R. 1985. New developments in
field and laboratory testing of soils. Proc. 11th Intl. Conf. on Soil Mech. & Fdn. Eng.,
Stockholm, Vol. 1, San Francisco, pp. 57-153.
Keane, A.J. 1995. Genetic algorithm optimization of multi-peak problems: studies in
convergence and robustness. Artif. Intell. Eng. 9(2), 7583.
Kim, S.K., 2008. Characterization of deltaic deposits in the Nakdong River mouth, Busan. Proc.
3rd Intl. Conf. on Site Characterization, Taipei, pp. 75-88.
Kim, Y.T., Lee,S.R.,1997. An equivalent model and back-analysis technique for modelling in

21

ACCEPTED MANUSCRIPT
situ consolidation behavior of drainage-installed soft deposits. Comput. Geotech. 20(2), 125
142.

PT

Leung, Y.-W., Wang.Y., 2001. An orthogonal genetic algorithm with quantization for global

RI

numerical optimization. IEEE T. Evolut. Comput. 5(1), 4153.

Park, H.I., Park, B., Kim, Y.T., Hwang, D.J., 2009. Settlement prediction in a vertical drainage-

SC

installed soft clay deposit using the Genetic Algorithm (GA) back analysis. Mar. Georesour.

NU

Geotec. 27(1), 17-33.

Renders, J.M., Flasse, S.P., 1996. Hybrid methods using genetic algorithms for global

MA

optimization. IEEE T. Syst. Man Cy. B.26(2): 243258.

Sakurai, S., Takeuchi, K., 1983. Back analysis of measured displacements of tunnels. Rock Mech.

TE

Rock Eng. 16(3), 173-180.

Shoji, M., Ohta, H., Arai, K., Matsumoto, T., Takahashi, T., 1990. Two dimensional

AC
CE
P

consolidation back-analysis. Soils Found. 30(2): 6078.

Tan, T., Inoue, T., Lee, S., 1991. Hyperbolic method for consolidation analysis. J. Geotech.
Engrg. 117(11), 1723-1737.

Tan, S.A., 1993. Ultimate settlement by hyperbolic plots for clays with vertical drains. J.
Geotech. Engrg. 119(5), 950-956.
Tan, S.-A., 1995. Validation of hyperbolic method for settlements in clays with vertical drains.
Soils Found. 35(1), 101-113.
Tan, S.-A., Chew, S.-H., 1996. Comparison of the hyperbolic and Asaoka observational method
of monitoring consolidation with vertical drains. Soils Found. 36(3), 31-42.
Terzaghi, K., Peck, R.B., Mesri, G., 1996. Soil Mechanics in Engineering Practice. John Wiley
and Sons, New York.

22

ACCEPTED MANUSCRIPT

1000

Probability of reproduction (%)

70 %

Probability of mutation (%)

0.5 %

Number of maximum generations

30

RI

Number of individuals

SC

Values

AC
CE
P

TE

MA

NU

Parameters

PT

Table 1. Basic parameters for optimization using the genetic algorithm.

23

ACCEPTED MANUSCRIPT
Table 2. Sectional information and back-analyzed consolidation parameters.

Upper
clay
layer

Lower
clay
layer

SP-1

30.0

25.0

SP-2

30.0

SP-3

Height of
surcharge
fill (m)

ch 10-3
(m2/day)

Cc
Lower
clay
layer
0.77

Upper
clay
layer
9.20

Lower
clay
layer
74.63

20.0

18.2

0.38

0.94

8.83

43.23

22.2

20.0

18.1

0.31

0.89

6.48

19.25

SP-4

22.5

20.0

19.9

0.31

1.05

9.13

14.72

SP-5

22.7

12.0

19.2

0.45

1.09

8.29

18.75

SP-6

22.0

10.0

16.3

0.45

0.96

37.57

6.79

SP-7

22.0

10.0

16.7

0.40

1.10

9.38

45.00

SP-8

18.5

14.9

0.45

18.15

SP-9

11.5

13.5

0.36

1.50

SP-10

8.5

10.9

0.18

16.81

SP-11

8.5

11.0

0.39

11.44

SP-12

23.0

7.0

18.0

0.58

0.71

10.52

98.36

SP-13

16.7

0.26

18.26

10.3

NU

MA
D

TE

AC
CE
P
-

SC

18.6

Upper
clay
layer
0.38

RI

Settlement
Plate

Back-analyzed parameters

PT

Thickness (m)

* Note: Cc (design) = 0.9 (upper clay layer), 0.6 (lower clay layer)
ch (design) = 8.5 10-3 m2/day (upper and lower clay layers)

24

ACCEPTED MANUSCRIPT

Design
stage

Hyperbolic
method

Asaokas
method

SP-1

Measured
final
settlement
(m)
5.19

6.84

5.56

5.14

5.35

SP-2

5.44

5.87

5.88

4.94

5.57

SP-3

5.96

6.51

6.50

5.62

6.11

SP-4

5.55

5.37

6.04

RI

Table 3. Predicted final settlements by settlement estimation methods.

5.62

5.58

SP-5

5.65

5.37

6.14

5.72

5.73

SP-6

4.21

5.17

4.57

4.19

4.33

SP-7

4.08

4.98

4.46

4.08

4.09

SP-8

2.74

3.77

2.94

2.70

2.69

SP-9

1.79

2.85

1.90

1.86

1.74

SP-10

0.79

2.21

0.85

0.94

0.71

SP-11

1.33

2.27

1.51

1.31

1.28

SP-12

4.80

6.02

5.18

4.85

4.92

SP-13

1.21

3.00

1.19

1.06

1.19

Predicted final settlement (m)

25

Genetic
Algorithm

PT

SC

NU
MA

TE

AC
CE
P

Settlement
Plate

ACCEPTED MANUSCRIPT
Fig. 1. Location of the studied site and sectional view of the deltaic deposits of the Nakdong
River (Modified after Kim (2008)): (a) Korean peninsula, (b) location of the studied site, and (c)

PT

sectional view of soil layers of the studied site.

RI

Fig. 2. Variation of soil properties on physical characteristics and compressibility with elevation.

SC

Fig. 3. Graphical settlement prediction methods: (a) hyperbolic method and (b) Asaoka method.

NU

Fig. 4. Flowchart of GA procedure.

MA

Fig. 5. Variation of back-analyzed consolidation parameter through the optimization process at

SP-6: (a) compression index and (b) horizontal coefficient of consolidation.

Fig. 6. Variation of back-analyzed consolidation parameter by elapsed time at SP-6: (a)

TE

compression index and (b) horizontal coefficient of consolidation.

AC
CE
P

Fig. 7. Settlement curves by consolidation theory, measurement, and back analysis at SP-3.

Fig. 8. Settlement prediction by the thickness of consolidation layer.

Fig. 9. Settlement curve by the GA back-analysis method in the case of modifying a surcharging
plan at SP-12: (a) original plan and measurement and (b) modified plan and measurement.

26

MA

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

AC
CE
P

TE

Fig. 1

27

MA

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

AC
CE
P

TE

Fig. 2

28

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

AC
CE
P

TE

MA

Fig. 3

29

ACCEPTED MANUSCRIPT

Generate initial population

PT

GEN = 0

RI

Calculate the objective function

SC

Calculate the fitness value

NU

Selection

Reproduction & Mutation

MA

Calculate the objective function

GEN=GEN+1

Fig. 4

AC
CE
P

TE

Yes

Calculate the fitness value

GEN < GENmax

No
Best individual
.

30

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

AC
CE
P

TE

MA

Fig. 5

31

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

AC
CE
P

TE

MA

Fig. 6

32

TE
AC
CE
P

Fig. 7

MA

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

33

MA

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

AC
CE
P

TE

Fig. 8

34

AC
CE
P

TE

MA

NU

SC

RI

PT

ACCEPTED MANUSCRIPT

Fig. 9

35