Missouri University of Science and Technology

Scholars' Mine
International Conferences on Recent Advances in 2010 - Fifth International Conference on Recent
Geotechnical Earthquake Engineering and Soil Advances in Geotechnical Earthquake Engineering
Dynamics and Soil Dynamics

May 24th - May 29th

Artificial Neural Network Model for Prediction of

Liquefaction Potential in Soil Deposits
F. Farrokhzad
Babol Noshirvani University of Technology, Iran

A. J. Choobbasti
Noshirvani University of Technology, Iran

A. Barari
Babol Noshirvani University of Technology, Iran

Follow this and additional works at: [Link]

Part of the Geotechnical Engineering Commons

Recommended Citation
Farrokhzad, F.; Choobbasti, A. J.; and Barari, A., "Artificial Neural Network Model for Prediction of Liquefaction Potential in Soil
Deposits" (2010). International Conferences on Recent Advances in Geotechnical Earthquake Engineering and Soil Dynamics. 4.
[Link]

This Article - Conference proceedings is brought to you for free and open access by Scholars' Mine. It has been accepted for inclusion in International
Conferences on Recent Advances in Geotechnical Earthquake Engineering and Soil Dynamics by an authorized administrator of Scholars' Mine. This
work is protected by U. S. Copyright Law. Unauthorized use including reproduction for redistribution requires the permission of the copyright holder.
For more information, please contact scholarsmine@[Link].
 ARTIFICIAL NEURAL NETWORK MODEL FOR PREDICTION OF
LIQUEFACTION POTENTIAL IN SOIL DEPOSITS
[Link] 2-A. J. Choobbasti [Link]
MS.c of geotechnical engineering, Associate professor, Departement MS.c of geotechnical engineering,
Departement of civil engineering, of civil engineering, Babol Departement of civil engineering,
Babol Noshirvani University of Noshirvani University of Babol Noshirvani University of
Technology, Babol, Iran. Technology ,Iran. Technology, Babol, Iran.
E-mail:
FarzadFarrokhzad2003@[Link]

ABSTRACT

With the increase in population, the evaluation of liquefaction is becoming more important for land use planning and development. In
soil deposits under undrained condition, earthquakes induce cyclic shear stresses, may lead to soil liquefaction. Artificial neural
network (ANN) is one of the, artificial intelligence (AI) approaches that can be classified as machine learning. Simplified methods
have been practiced by researchers to assess nonlinear liquefaction potential of soil. In order to address the collective knowledge built
up in conventional liquefaction engineering, an alternative general regression neural network model is proposed in this paper.

To meet this objective, a total of 30 boreholes are introduced into the model. The data includes the results of field test from (Babol,
Mazandaran, Iran).

The results produced by the proposed Artificial Neural Network model compared well with the determined liquefaction decision
obtained by simplified methods. It provides a viable liquefaction potential assessment tool that assist geotechnical engineers in making
an accurate and realistic predictions. Furthermore, this study integrates knowledge learned from field test and seismic parameters to
the ongoing development of liquefaction analysis.

The results show that there is liquefaction potential in western part of Babol, and in southern part of Babol no liquefaction potential
were seen. In middle part and eastern part low liquefaction potential were predicted by ANNs. This study shows that neural networks
are a powerful computational tool which can analyze the complex relationship between soil liquefaction potential and effective
parameters in liquefaction.

INTRODUCTION

When saturated sand deposits are subjected to earthquake- Liquefaction is a phenomenon in which the strength and
induced shaking, pore water pressures are built-up leading to stiffness of a soil is reduced by earthquake shaking or other
liquefaction or loss of soil strength. Major earthquakes that rapid loading. During the liquefaction, pore water presure
have occurred during past years, such as the 1964 Alaska, exerts a pressure on the soil particles that influences how
1964 Niigata, 1989 Loma-prieta and the 1995 Hyogoken- tightly the particles themselves are pressed together. Prior to
Nambu have demonstrated the damaging effects of soil an earthquake, the water pressure is relatively low (Ishihara K,
liquefaction. Therefore, it is necessary to obtain a proper Yasuda S). However, earthquake shaking can cause the water
understanding of effective parameters such as soil properties pressure to increase to the point where the soil particles can
and nature of earthquake on severity of soil liquefaction (Seed readily move with respect to each [Link] shaking
HB, Idriss IM, Makdisi F, Banerjee N). often triggers this increase in water pressure, but construction
related activities such as blasting can also cause an increase in
water pressure.

Paper No. 4.46a 1

When liquefaction occurs, the strength of the soil decreases liquefaction. Medium dense soils initially showed the same
and, the ability of a soil deposit to support foundations for behavior as the loose samples but, after initially exhibiting
buildings and bridges is reduced (Seed HB, Idriss IM). contractive behavior, the soil transformed and began
exhibiting dilative behavior. Castro referred to this type of
In the 1960, Gonzalo Castro, a student of Casagrande, behavior as limited liquefaction (Whitman RV).
performed an important series of undrained, stress-controlled
triaxial tests. Castro observed three different types of stress-
strain behavior depending upon the soil state. Dense
specimens initially contracted but then dilated with increasing
effective confining pressure and shear stress. Very loose
samples collapsed at a small shear strain level and failed
rapidly with large strains. Castro called this behavior
liquefaction; it is also commonly referred to as flow

Fig. [Link] triaxial test stress paths for two specimens of different densities.

Ground response analyses based on the finite element method from training data. An ANN consists of several layers of
provide a better assessment of liquefaction of a soil deposit by highly interconnected computational units called neurons.
taking into account the nature of the earthquake and the pore Figure 2 shows the general structure of a three layer feed-
pressure dissipation; they are often costly and time consuming. forward ANN. The neural network contains one input layer,
In addition, constitutive models used in those programs need one or two hidden layers, and one output layer The number of
large number of parameters to determine the pore pressure nodes in the input layer equals the number of parameters in the
generation in soil due to earthquake loading. Therefore, process. The output layer represents the quality responses of
simplified methods in assessing soil liquefaction are popular the product (Agrawal, G., Weeraratne, S., and Khilnani, K).
among practicing engineers. These procedures are very useful The hidden layer represents the interactions between the input
at the preliminary design stages to assess the liquefaction risk. and output layers. Normally the number of nodes in the hidden
If the liquefaction risk is high, then a detailed finite element layer is set to be half of the total number of input nodes and
analysis can be carried out to obtain the pore pressure output nodes. If the relationships between the operation
distribution and ground displacement along the depth of the parameters and quality responses are difficult to identify, two
soil deposit, which is necessary in subsequent design of deep hidden layers may be used. Such neural networks are capable
foundations. In more details improving the reliability of of capturing complex nonlinear relationships inherent in a
liquefaction risk, may lead to cost reduction and helps to process (Hornik K).
operation planning (NCEER).
The ANN uses a set of examples in a training database as
An artificial neural network is a mathematical model or input, a learning algorithm to adjust the weights and an
computational model based on Biological neural networks. It activation function to derive an output. If the connection
consists of an interconnected group of artificial neurons and weight between the neurons is changed, the relationship of the
processes information using a connectionist approach to network’s output to its input will be altered. The process of
computation. In most cases an ANN is an adaptive system that adjusting the connection weights by repeatedly exposing the
changes its structure based on external or internal information network to known input-output data is called training. The
that flows through the network during the learning phase. error back-propagation learning method is the most popular
and successful training technique. A trained ANN can take
Artificial neural networks mimic human brains to learn the inputs and produce outputs very quickly, which is an
relationships between certain inputs and outputs from advantage in doing optimization in the proposed approach
experience. They are considered as information processing (Agrawal, G., Chameau, J. A., and Bourdeau, P. L).
systems that have the abilities to learn, recall and generalize

Paper No. 4.46a 2

ANNs have been proved to be an universal estimator, hence setting. In this research, the ANN is used to establish the
they are promising techniques in solving pattern recognition nonlinear multivariate relationships between liquefaction
and classification, optimization and function approximation potential and parameters, which can be used to predict the
problems. Recently, ANNs are used to model complex liquefaction potential in soil.
manufacturing processes and to identify the optimal process
Ellis et al. (1995) developed an ANN model for sands based
on grain size distribution and stress history. Sidarta and
Ghaboussi (1998) employed an ANN model within a finite
element analysis to extract the geometerial constitutive
behaviour from non-uniform material tests. Penumadu and
Jean-Lou (1997) used neural networks for representing the
behaviour of sand and clay soils. Ghaboussi and Sidarta
(1998) used neural networks to model both the drained and
undrained behaviour of sandy soil subjected to triaxial
compression-type testing. Penumadu and Zhao (1999) also
used ANNs to model the stress-strain and volume change
behaviour of sand and gravel under drained triaxial
compression test conditions. Zhu et al. (1998a; 1998b) used
Fig. 2. A three-layer feed-forward neural network structure. neural networks for modelling the shearing behaviour of a
fine-grained residual soil, dune sand and Hawaiian volcanic
Recently, extensive studies have been done on application of soil (Malvić, T., Velić, J. And Cvetković).
ANN to Geotechnical engineering problems. Chan et al.
(1995) developed a neural network as an alternative to pile It is known, that the engineering properties of soil varied from
driving formulae. The network was trained with the same point to point and uncertain behaviour due to the complex and
input parameters listed in the simplified Hiley formula (Broms partially predictable physical processes associated with the
and Lim 1988), including the elastic compression of the pile forming of these deposits. This is in contrast to most other
and soil, the pile set and the driving energy delivered to the civil engineering materials, such as steel, concrete and timber,
pile (Abu-Kiefa, M. A). which exhibit far greater homogeneity and isotropy. In order
to cope with the complexity of geotechnical behaviour, and
Lee (1996) utilized neural networks to predict the ultimate the spatial variability of soil deposits, traditional forms of
bearing capacity of piles. The problem was simulated using engineering design models are justifiably simplified. It is also
data obtained from model pile load tests using a calibration known, that assessing liquefaction potential of soil plays an
chamber and results of insitu pile load tests. Teh et al. (1997) important role in geotechnical evaluation for construction of
proposed a neural network for estimating the static pile major structures (Cvetković).
capacity determined from dynamic stress-wave data for
precast reinforced concrete piles with a square section. Several methods for liquefaction assessment have been
developed. One method of analyses (Seed and Idriss) proposes
Sivakugan et al. (1998) explored the possibility of using using the estimated shear stress level and cycle number likely
neural networks to predict the settlement of shallow to be developed in the field, due to a design earthquake.
foundations on granular soils. A neural network was trained Comparison of these stresses with those causing liquefaction
with five inputs representing the net applied pressure, average of soil samples obtained from laboratory tests helps
blow count from the standard penetration test, width of identifying the liquefiable zones of a deposit. Another method
foundation, shape of foundation and depth of foundation. The (Seed et al.) considers field observations of performance of
output was the settlement of the foundation (Riedmiller, M. sites during previous earthquakes. By combining the data on
and Braun, H). earthquake characteristics and insitu properties of soil
deposits, an empirical relationship is established.
Most recently, Shahin et al. (2000) carried out similar work
for predicting the settlement of shallow foundations on The purpose of this research is to investigate the effect of the
cohesionless soils. In this work, 272 data records were used soil and seismic parameters, with an artificial intelligence
for modelling. The input variables considered to have the most computational tool, and its success in assessing liquefaction
significant impact on settlement prediction were the footing potential (National Research Council).
width, the footing length, the applied pressure of the footing
and the soil compressibility (). The results of the ANN were Data collection in explored soils is important for assessing of
compared with three of the most commonly used traditional liquefaction as well as estimation of strata thickness, soil type,
methods. These methods were Meyerhof (1965), Schultze and groundwater table etc. It is also time consuming and often
Sherif (1973) and Schmertmann et al. (1978). The results of expensive process, which includes many field and laboratory
the study confirmed those found by Sivakugan et al. (1998), in experiments. Therefore reliable prediction of liquefaction asks
the sense that ANNs were able to predict the settlement well for carefully planning of sampling, testing and exploration
and outperform the traditional methods (Cal, Y). methods. Data had been collected from the boreholes

Paper No. 4.46a 3

(maximum depth: 30 m) over a 6 square kilometres area of
Babol municipal region. Artificial neural networks are trained
with 60% and validated with 10% of borehole data for
prediction of liquefaction. The whole system is eventually
tested for efficiency, using 30% of borehole data left for test
of the network, distributed randomly over the study area.
Based on the obtained results and considering that the test data
were not presented to the network in the training and
validation process, it can be stated that the trained neural
networks are capable of predicting variations in the
liquefaction potential of soil with an acceptable level of
confidence (Malvić, T. And Prskalo, S).

Successful prediction of liquefaction in soil deposit using the

existing data leads to improve the reliability of data which will
be used for construction in future. Such approach is presented
in the following text that generally comprises presentation of
the study area, then description and selection of the neural
model, its training, improving, and developing of final model Fig. 3. Map of study area (top) and the zone of the Babolrood
used for prediction of liquefaction by specific ANN (Agrawal, river (bottom).
G., Weeraratne, S., and Khilnani, K). Very often in geotechnical engineering, it is possible to
encounter some types of problems that are very complex and
MATERIALS AND METHODS can not be completely understood. Mathematical models that
attempt to solve such problems can not included entire physics
Babol, a city of Mazandaran province in the northern part of of process and necessarily need to simplify the model or
Iran, is considered as the study area. As shown in Figure 3 the incorporating some assumptions. Mathematical models also
city is located approximately 20 kilometres south of Caspian assumed the knowing of model structure in advance, which
sea on the west bank of the river Babolrood and receives does not need to be optimal. Consequently, many
abundant annual rainfall. Babolrood has 2 groups of river mathematical models fail to simulate the complex behavior of
terraces, namely H1 and H2. H1 is referred to as river terraces most geotechnical engineering problems. In contrast, ANNs
with down surface level of height one to 2.5 (m) and width of are based mostly on the input data structure, assuming that
0 to150 m. It is as boundary of active (yearly) flood plain in such structure and interaction among data can describe the
parts of river and it is as alternative flood plain in many prediction model. In this case, there is no need to neither
sections. It consists of fine-grained and unconsolidated simplify the problem nor incorporate any assumptions (expect
alluvial sediments. H2 is referred to as river terraces; with user selection of data that are in some meaningful connection).
high surface level of 4-6 (m). Vegetation on surface of terrace Moreover, one obtained neural models can always be again
is compact. It consists of materials of Aeolian deposits (i.e. trained with more extensive and newer dataset from the same
loess). Most major earthquakes occur around the boundaries of area with goal to reach better results.
the tectonic plates such as those that exist in north of
Iran. The data used in presented research, includes borehole logs
(data collected from digging boreholes) bored in the study area
(Figure 4) and is collected by different institutions for
different research purposes. The database includes more than
40 borehole logs in an area of more than 6 square kilometres
from Babol zone.

Paper No. 4.46a 4

 connection weights or weighting coefficients (like in
interpolation methods), usually marked as wij1, Also the
training and validation sets are used during the training
process and the test set is used for obtaining the estimates. All
ANN models was trained using the automated regularization
algorithm to improve generalization. The validation set served
as a constraint on training, in order to minimize over fitting.

The usefulness of the neural network approach for populating

the similarity model is presented In this case study. The inputs
to the network were data on a set of soil formative
environmental factors; the output from the network was a set
of similarity values to a set of prescribed soil classes divided
by grain size, thickness of each layer and groundwater table. A
set of 2500 samplings are performed in study area from 30
boreholes. Data are collected using geotechnical investigation.
Each sample is carefully checked, because to ensure the
accurate prediction of an ANN model we need to build a
Training borehole …. reliable training, validating and testing sets.

Validation borehole … In this analysis, regarding the available data and their quality,
a neural network program written in back propagation
Testing borehole ……. algorithm, is used. Eight soil and seismic parameters are
selected as input in different models, and these parameters are
Fig. 4. The 6 zones in Babol area. divided into data groups. Each data group is introduced to the
network individually, and performance of the network on the
From the total of 40 raw borehole data, only 30 logs with a assessment of liquefaction potential is investigated. The
depth range of 10 to 30 meters were acceptable for using in network predictions are compared with the conventional
ANN. The regular tests were performed on the samples. liquefaction determination method proposed by Seed et al.

The available data set is divided into three sets, namely Back propagation is selected as the training algorithm of
training, validation, and test sets, based on random selection. neural network (Table 2). It is the best known training
This way we can examine the validity of the model in a more algorithm for multilayer perceptrons neural networks, and still
comprehensive manner (Choobbasti AJ, Farrokhzad F, Barari one of the most useful and later improved in some advanced
A). In ANN forecasting models, 60% of the records are forms like RProp. Back propagation algorithm means that
selected as training, 30% are taken for test for final evaluation, network training includes determination of the difference
and the remaining 10% used for validation or monitoring the between true and wanted network response, i.e. means
performance of the model during the training phase (Table 1). calculation of error that is backed in the neural network for
obtaining optimal training. It has lower memory requirements
Table 1. Performance of different sets of data used in ANN. than most algorithms, and usually reaches an acceptable
estimation error quite quickly (in relative low number of
Training Validation Testing iterations or epochs).
set set set
Number of boreholes 18 3 9 The ANN model for this study was developed, trained,
validated and tested within STATISTICA computational
Number of data 1500 250 750
environment utilizing the neural network toolbox. And the
(I/O data pairs)
accuracy of the ANN model was evaluated using RMSE
between measured and predicted values and pressed as:
In problems dealing with different variables and with different n
ranges and dimensions, the application of several networks
may be a good choice. Neural networks are efficient tools  (z s  z0 ) 2
when used as pattern classifiers, it is important to properly RMSE  k 1

select the input variables for training (learning) process of n

ANNs, as the way how to determine relationships between (1)
input and output variables. A set of known input and output
values is named as input-output pair. All such pairs are usually
divided into three sets. The first and second are described as nk 1

training and validation sets which are used to determine the

1
- y  f ( y k)  f (
k
i
'
i w
j 1
k
ij yik 1 )

Paper No. 4.46a 5

Where zs is observed value, z0 is predicted value, n is instant.
number of samples. The RMSE of the different neurons in
hidden layer is plotted in Figure 5. The ANN architecture for
prediction of soil classification and layers thickness in the
study area was a feed forward, supervised, multilayer
perceptron (MLP) network with one hidden layer and an
output layer. The best fitting training data set was obtained
with six neurons in the hidden layer for prediction of
liquefaction.

In the selection of learning / training algorithm number of

neurons in different layers (input, hidden, output), number of
epochs, learning rate and the momentum have been applied

Table 2. Results of research in order to Learning / training algorithm selection.

Supervised Back propagation Conjugate Levenberg- Quick Delta

Learning/ Gradient Marquardt Propagation -bar-
training Descent Delta
algorithms
RMSE (%) 6.3 12.1 8.7 10 9.2
([Link])

12

10

8
RMSE

2
(min. error)
0
0 2 4 6 8 10
Number of neurons in hidden layer

Fig. 5. The RMSE of the different neurons in hidden layer

for prediction of soil liquefaction potential.

In each epoch, the entire training set is fed through the layer network is sufficient to uniformly approximate any
network, and used to adjust the network weights. Numbers of continuous and nonlinear function. The model architecture
epochs are specified at the start, but also alternative stopping was built with one hidden layer, a learning rate of 0.1 updated
criterion may also be specified, and if over-trained network with a coefficient of 1.1 after each epoch and a momentum
occurs the best network discovered during training can be term of 0.9 updated with a coefficient of 0.95 after each
retrieved. In this analysis, the number of epochs varied epoch. The input vector is fully connected to the hidden
between 100 and 400. neurons by a tan-sigmoid transfer function and the neurons of
hidden layer are fully connected to the output layer via a linear
A batch mode feed-forward multilayer perceptron (MLP) with function. Experimental studies were started with one hidden
back-propagation learning rules was used to create the desired neurons to reach the optimum number of hidden neurons and
ANN model using STATISTICA software. Also, an adaptive desired precision. Input vector contains soil initial parameters
learning rate was employed to keep the learning step size as and output (the target vector) is liquefaction potential. In order
large as possible while the training is stable. According to a to obtain a more efficient training process, the input and target
universal approximation theorem, demonstrated concurrently were standardized to have zero mean and unity standard
by several researchers for traditional MLP, a single hidden deviation. Cross-validation or employing another set of data

Paper No. 4.46a 6

for more testing can be used to increase the generality of the
models for future predictions. In this study, 10% of borehole RESULTS AND DISCUSSION
data were used as validation set. In fact, several ANN models
using element tests data were constituted for generating the
models. Among them, the model with better performance In the previous section, the learning or training dataset is used
(greater coefficient of determination and smaller RMSE) to determine the weights. Then a second validation set is used
for validation data set was selected. In other words, the to monitor the performance of the model during the training
ANN models were developed with the best performance phase and to minimize over fitting and finally the test sets to
concurrently for training, testing and validation data sets. evaluate the trained neural network. It is evident from test data
Three different ANN models were developed using different sets that the experimental ANN can be applied successfully to
combinations of input parameters in Table 3. predict liquefaction potential.
It can be seen from Table 3 that, except for model #1, The samples are divided in to 3 groups (training, validation
performances of the models are generally improved when and testing). In Figure 6 samples of testing group are
input parameters are increased. correlated in terms of sample number and the accuracy
(comparison between prediction and real data) of each sample
is shown. In these figures, terms of the ratio of actual data per
predicted value (in Y-axis) versus Case number (in X-axis) for
Table . 3. Different combinations of input parameters. different soil samples are presented. It is clear that if the
predicted and the true values were the same, such point lie on
Model # 1 2 3 line y=1. Scattering pattern indicates on differences. It is clear
Input
M,
a , M,
 , Vs ,  ,
that the average correlation of the ANN model and true data in
all case is over 90%. So it can be concluded, that the
g   prediction of liquefaction potential agrees with calculated
 ,   , Rd , Dr Vs , Rd ,  Rd ,  value collected from boreholes.
RMSE 13% 17% 16%

1.2 7
Calculated value/predicted value

6
FOS (Prediction of ANN)
1
5
0.8
4
0.6
3
0.4
2

0.2 1

0 0
0 50 100 150 200 250 0 1 2 3 4 5 6

Sample number FOS (Calculated Value with seed and Idress method)

1.2 7
Calculated value/predicted value

6
FOS (Prediction of ANN)

1
5
0.8
4
0.6
3
0.4 2

0.2 1

0 0
250 300 350 400 450 500 0 1 2 3 4 5 6

Sample number FOS (Calculated Value with seed and Idress method)

1.2 6
Calculated value/predicted value

FOS (Prediction of ANN)

1 5

0.8 4

0.6 3

0.4 2

0.2 1

0 0
500 550 600 650 700 750 0 1 2 3 4 5 6

Sample number FOS (Calculated Value with seed and Idress method)

Fig. 6. Errors involved ANN for prediction of liquefaction potential.

CONCLUSION

In this research, the data used to train the model were taken boreholes. The average accuracy between the ANN prediction
2
and real data in all cases is over 90%. The liquefaction
from area of 6 km of Babol region in the northern Iran. The
potential of a soil mass during an earthquake is dependent on
dataset encompasses 2500 sampling points (samples) from 30
both seismic and soil parameters. The impact of these soil and

Paper No. 4.46a 7

seismic variables on the liquefaction potential of soil is Ishihara K, Yasuda S. (1975). “Sand liquefaction in hollow
investigated through computational and knowledge based cylinder torsion under irregular excitation”. Soils and Found
tools called neural networks. A back-propagation neural 1975; 15(1):45–59.
network model is utilized. The back propagation learning
algorithm is a developing computational technique that assists Choobbasti AJ, Farrokhzad F, Barari A. (2009) “Prediction of
in the evaluation of experimental and field data. The artificial slop stability using artificial neural network”. Arab J Geosci;
neural network is trained using actual field soil records. The 10.1007/s12517-009-0035-3.
performance of the network models is investigated by
changing the soil and seismic variables including earthquake National Research Council. (1985). “Liquefaction of soils
magnitude, initial confining pressure, seismic coefficient, during earthquakes”. National research Council Report CETS-
relative density, shear modulus, friction angle, shear wave EE-001, National Academy Press, Washington, DC.
velocity and electrical characteristics of the soil. The most
efficient and global model for assessing liquefaction potential NCEER. (1997). In: Proceedings of the NCEER workshop on
and the most significant input parameters affecting evaluation of liquefaction resistance of soils. National Center
liquefaction are summarized. A forecast study is performed for for Earthquake Engineering Research. Technical Report No.
the city of Babol, Iran. NCEER970022, State University of New York at Buffalo,
New York, December 31.
Based on the obtained results, it can be stated that the trained
neural networks are capable of predicting liquefaction Cal, Y. (1995). “Soil classification by neural-network.”
potential with an acceptable level of confidence. It is believed Advances in Engineering Software, 22(2), 9597.
that, the prediction of liquefaction potential is a complex area
of research requiring detailed investigation also with other Abu-Kiefa, M. A. (1998), “General regression neural networks
methods, fieldwork and laboratory experiments. Further work for driven piles in cohesionless soils.” J. Geotech. &
on presented topic would be very useful to modify the Geoenv. Engrg., ASCE, 124(12), 1177-1185.
procedure for better adapting artificial neural network with
concept of prediction of liquefaction potential. Agrawal, G., Chameau, J. A., and Bourdeau, P. L. (1997).
“Assessing the liquefaction susceptibility at a site based on
information from penetration testing.” In: Artificial neural
REFERENCES networks for civil engineers: fundamentals and applications,
N. Kartam, I. Flood, and J. H. Garrett, eds., New York, 185-
Seed HB, Idriss IM. (1971) “Simplified procedure for 214.
evaluating soil liquefaction potential”. J Soil Mech Found
Div., ASCE; 97(SM8):1249–74. Malvić, T., Velić, J. And Cvetković, M. (2008): “Review of
neural network analyses performed in Croatian part of
Hornik K. (1991). “Approximation capability of multilayer Pannonian basin (petroleum geology data)”. XII. Congress of
feedforward networks as universal approximators”. Neural Hungarian geomathematics and the first congress of Croatian
Networks; 4: 251–7. and Hungarian geomathematics, Morahalom, Hungary, 29-
31.05.2008.
Whitman RV. (1971). “Resistance of soil to liquefaction and
settlement”. Soils Found 1971; 11(4):59–68. Malvić, T. And Prskalo, S. (2007): “Some benefits of the
neural approach in porosity prediction (Case study from
Cvetković, M. (2007): “Petroleum Geology Use of Neural Beničanci field)”. Nafta, 58, 9, 455-467. – generally about
Networks of the Example of Reservoir in Kloštar Field.” porosity mapping in Neogene breccia.
Unpublished Diploma thesis (in Croatian), Faculty of Mining,
geology and Petroleum Engineering, University of Zagreb, Chan, W. T., Chow, y. K., and Liu, L. F. (1995). “Neural
date of presentation 15.6.2007. (mentor Prof. Dr. Josipa Velić, network: An alternative to pile driving formulas.” J.
co-mentor: Dr. Tomislav Malvić), 49 p. – generally about Computers and Geotechnics, 17, 135-156.
prediction of sandstone/marls and saturation with oil/water in
Neogene sandstone. Agrawal, G., Weeraratne, S., and Khilnani, K. (1994).
“Estimating clay liner and cover permeability using
Seed HB, Idriss IM, Makdisi F, Banerjee N. (1975). computational neural networks.” Proc., First Congress on
“Representation of irregular stress time histories by equivalent Computing in Civil Engrg., Washington, June 20-22.
uniform stress series in liquefaction analyses”. Report No.
UCB/EERC-75/29, Earthquake Engineering Research Centre, Riedmiller, M. and Braun, H. (1993): “A direct adaptive
U.C. Berkeley. method for faster backpropagation learning“: The Rprop
algorithm. Proc. Of the IEEE Int. Conf. On Neural Networks,
San Francisco, p. 586-591.

Paper No. 4.46a 8