Hindawi

Advances in Civil Engineering

Volume 2020, Article ID 2569531, 11 pages
https://doi.org/10.1155/2020/2569531

Research Article
Assessment of Waterlogging Risk in the Deep Foundation Pit
Projects Based on Projection Pursuit Model

Han Wu and Junwu Wang

School of Civil Engineering and Architecture, Wuhan University of Technology, Wuhan 430070, China

Correspondence should be addressed to Junwu Wang; [email protected]

Received 5 February 2020; Accepted 15 May 2020; Published 29 May 2020

Academic Editor: Qiusong Chen

Copyright © 2020 Han Wu and Junwu Wang. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
As a result of global climate change and urbanization, waterlogging disasters have occurred frequently around the world, and deep
foundation pit projects with lower terrain suffer even more. This study puts forward a method for the waterlogging risk assessment
of deep foundation pit projects via the combination of a projection pursuit model, particle swarm optimization, and an in-
terpolation algorithm. First, through a comprehensive analysis of the water circulation process in waterlogging and the char-
acteristics of deep foundation pit projects, a risk index system with 11 indicators is identified and constructed. Then, a projection
pursuit model optimized by particle swarm optimization is leveraged to determine the weights of the indicators and the best
projection values of evaluation objects, and the mathematical function between the best projection values and the risk levels is
constructed by an interpolation algorithm. Finally, three deep foundation pit projects of the Chengdu Metro Line 11 in China are
selected as case studies. The results demonstrate that the frequency of storms, intensity of rainfall, preparation of emergency rescue
plans, and proportion of older workers have the greatest impacts on waterlogging risk in deep foundation pits. The risk ranking of
the case studies is found to be consistent with the actual situations, which proves the objectivity and effectiveness of the
proposed method.

1. Introduction for disaster prevention, mitigation, and preparedness work

in the project management of deep foundation pit
In recent years, under the coupled effects of global climate engineering.
change and urbanization, urban rainstorms and floods have The assessment manner of waterlogging risk is to rate the
frequently occurred and pose tremendous threats to public level of risk via an established mathematical model and
safety [1, 2]. In July 2016, Wuhan, the capital city of Hubei evaluation index. According to the key influencing factors of
Province in China, suffered a serious waterlogging disaster. waterlogging disaster risk and different risk levels, project
This catastrophe caused a direct economic loss of 351 million managers are able to implement different measures of
RMB, and transport and traffic were paralyzed [3]. In waterlogging prevention, mitigation, and preparedness.
January 2020, heavy rain and floods in Jakarta, the capital of Scholars have carried out substantial research on urban
Indonesia, resulted in the death of 16 people and the waterlogging risk assessment. Hou [4] constructed a
evacuation of 35,600 people. The construction sites of deep stormwater management model by combining geographic
foundation pit projects are extremely low; rainwater information system (GIS) and remote sensing (RS) tech-
therefore naturally converges into these pits, making them nologies. This model was used to simulate the changes of the
more vulnerable to waterlogging disasters. Thus, conducting inundation range and accumulated water depth with the
the waterlogging risk assessment of deep foundation pit increase of rainfall intensity in the large-scale area of Jinfeng
projects will alleviate the injuries and property losses caused District, Yinchuan City, China. From the perspective of the
by waterlogging. This is of the highest guiding significance discipline of disaster science, Jia et al. [5] studied the risk
2 Advances in Civil Engineering

level of waterlogging in prefecture-level cities in Henan, determine the optimum projection direction, which is a
China. The results showed that the ability to prevent and complex nonlinear optimization problem [17]. At present,
reduce disasters had a critical impact on rainfall and most researchers adopt the GA to solve this problem, though
waterlogging risk. Wu et al. [6] investigated the vulnerability it is characterized by some shortcomings including the
of Zhengzhou City to flood disasters from the perspective of dependence on the initial population selection, a slow
government management. Yu et al. [7] used an optimal convergence speed, and too many parameter settings
weighting method and fuzzy comprehensive evaluation [18, 19]. Compared with a GA and ant colony optimization,
method to study waterlogging risk assessment in the con- particle swarm optimization (PSO), which is also a typical
struction and operation of subway stations; however, this metaheuristic algorithm, is characterized by fast conver-
indicator system was too simple and was unable both to gence and fewer parameter settings [20]. Although PSO does
consider the water circulation process during waterlogging not require tedious mathematical operations, it has been
and to effectively reflect the characteristics of subway sta- proven to better solve most optimizations. Dormishi et al.
tions. In addition, this study did not incorporate high-di- [21] used a variety of metaheuristic algorithms to com-
mensional data that could be obtained before evaluation, prehensively evaluate cutting machine performance in the
which reduces the application and popularization values of mineral processing field, and the research results demon-
the proposed model. In summary, existing related research strated that the PSO exhibited advantages over a differential
has been primarily focused on large-scale areas such as cities evolution algorithm. Hasanipanah et al. [22] used the PSO
and provinces, rather than small-scale areas such as projects model and the other methods to, respectively, predict the
under construction. To the best of the authors’ knowledge, ground vibration caused by an explosion; the results also
research on the assessment of the waterlogging risk in deep revealed that the PSO model performed better than the other
foundation pits has not yet been reported. methods.
Currently, soft computing techniques are widely used in According to the preceding analysis, the present paper
risk assessment and have achieved excellent research results. proposes a method of waterlogging risk assessment using the
Mikael et al. [8] used the harmony search algorithm to ef- combination of a PPM and PSO. The main contributions of
fectively evaluate the geological disasters of railway tunnels. this paper are as follows. (1) From the perspectives of the
Bui et al. [9] studied the flood forecasting by the multivariate fields of environmental science and civil engineering, a
adaptive regression splits and PSO. This paper pointed out rainstorm waterlogging disaster risk index system that can
that PSO had better reliability and accuracy than other ar- reflect the characteristics of small-scale areas in deep
tificial intelligence algorithms. Artificial bee colony algorithm foundation pit engineering and construction project man-
was also used to evaluate the risks of tunnel projects [10]. agement is constructed for the first time. (2) The PPM is used
There are numerous factors that affect the waterlogging to effectively deal with the problem of the high-dimensional
risk of deep foundation pits, and waterlogging risk assess- data of rainstorm waterlogging disaster risk, and PSO is used
ment data is therefore high-dimensional. Determining how to determine the optimum projection direction of the PPM.
to effectively deal with this high-dimensional data is the key (3) This paper, for the first time, reveals that the frequency of
for improving the objectivity and effectiveness of water- storms, intensity of rainfall, preparation of emergency rescue
logging risk assessment. The projection pursuit model plans, and proportion of older workers have the greatest
(PPM) projects high-dimensional data into low-dimensional impacts on the waterlogging risk of deep foundation pits and
space for analysis. In recent years, it has been used in- are the key factors in project management. The series of
creasingly more in the field of risk evaluation [11] and research in this paper provides a scientific basis for the
decision-making [12] to effectively evaluate high-dimen- prevention, mitigation, and preparedness of waterlogging
sional data. disasters in deep foundation pit engineering.
To effectively analyse complex indexes in the research of The remainder of this paper is organized as follows.
gas outburst prediction, Liang [13] established a prediction Section 2 details the research materials and methods, in-
model by using a PPM optimized by a genetic algorithm cluding index selection and the risk assessment method. The
(GA). In this model, the one-dimensional projection values process of the case analysis and discussion of the results are
calculated by the GA were used to indicate the potential gas presented in Section 3, and the research conclusions are
outburst risk, and the PPM was confirmed to be objective presented in Section 4.
and effective. Liu et al. [14] proposed a PPM improved by the
Ameliorative Moth-Flame Optimization (AMFO) algorithm
for high-dimensional data for the evaluation and spatial
2. Materials and Methods
change rule of surface water quality. The empirical analysis 2.1. The Index System of Waterlogging Risk in Deep
results showed that the proposed AMFO-PPE was stable and Foundation Pits
reliable. To effectively evaluate the sustainability of pre-
fabricated parts, Jiang et al. [15] used a PPM optimized by 2.1.1. Determination of Risk Factors of Waterlogging Risk in
the Real-code Accelerated Genetic Algorithm (RAGA), Deep Foundation Pits. The risk assessment of waterlogging
based on which a customer satisfaction evaluation algorithm in deep foundation pits is a complicated and interdisci-
for an e-commerce platform was constructed [16]. plinary issue. In the field of environmental science, water-
According to the previous research results, the key step in logging disasters are always provincial; thus, waterlogging
the processing of high-dimensional data by a PPM is to risk factors should be considered from the perspective of the
Advances in Civil Engineering 3

water cycle process of waterlogging disasters in both deep research and the review of local water resources an-
foundation pits and their surrounding areas [23]. The factors nouncements, local yearbooks, etc. Considering the un-
associated with the surrounding area reflect the characteristics availability of partial data in current engineering practice,
of waterlogging disasters in large-scale urban areas, while the X4, X8, X9, and X11 are all qualitative indexes without
factors associated with deep foundation pits reflect the de- measurement units, and their scores were obtained by a
tailed process of waterlogging disasters. The factors that questionnaire survey. X1, X2, X3, and X7 are benefit-based
clearly indicate the characteristics of deep foundation pit indicators; the greater the numerical values of these indi-
engineering should then be selected from the perspective of cators, the greater the level of waterlogging risk. The
civil engineering; these factors should naturally reflect both remaining indicators are cost-based; the smaller the nu-
structural characteristics and project management charac- merical values, the greater the level of waterlogging risk.
teristics. Finally, the factors related to the capacity of disaster
prevention and mitigation should be chosen from the per- 2.1.3. Assessment Standard of Waterlogging Risk.
spective of scientific fields such as disaster management. Currently, there are no uniform standards for the assessment
In the project management practice of deep foundation of waterlogging risk [28, 29]. To meet the needs of deep
pit engineering, many construction and management mea-
foundation pit project management practice, the risk as-
sures are used to resist waterlogging. For example, a retaining
sessment levels in this study were classified by the different
wall along the foundation pit is usually put in place to prevent
risk response measures that should be taken. Four levels of
waterlogging; the higher the height of the retaining wall is, the waterlogging risk, namely, low risk (I), moderate risk (II),
harder the floods to enter the construction site are, and the high risk (III), and extreme risk (IV), were identified. The
less risk of waterlogging there is. Water pumps are also often low risk (I) category denotes that there is no need to take
stored and utilized for draining in deep foundation pits; the further measures, only to check how existing measures are
more the pumps there are, the faster the water is pumped out implemented. Moderate risk (II) indicates that additional
and less the risk of waterlogging there is. In addition, the measures should be developed to deal with the risk of
waterlogging risk in deep foundation pit projects is related to waterlogging. High risk (III) implies that there is no need to
the population structure [24]; the public emergency capacity stop the construction operation, but further measures must
of citizens over 50, namely, the elderly, is significantly hin-
be immediately formulated to reduce the level of water-
dered; thus, these populations are more vulnerable when
logging risk as soon as possible. Extreme risk (IV) signifies
faced with emergencies. In other words, the greater the
that the construction task must be suspended immediately
proportion of construction workers over the age of 50, the
and that project managers are obliged to promptly take
greater the vulnerability the society has to waterlogging di- measures to decrease the waterlogging risk level.
sasters [25] and the higher the risk of waterlogging in deep As indicated in Table 1, the standards of different eval-
foundation pit projects. uation levels for the 11 indicators were established by com-
Referring to previous research results [26, 27], the risk bining the expert experience of project managers and previous
factors associated with disaster prevention and mitigation research results [30, 31]. The risk level descriptions of the four
are the preparation of an emergency rescue plan, efficiency qualitative indicators, X4, X8, X9, and X11, are divided into 2
of emergency rescue, proportion of relief workers, and re- components, namely, the qualitative language description and
serve of emergency rescue materials. An emergency rescue
corresponding quantitative score range. For example, the low
plan is the foundation for emergency management when
risk (I) level of X8 is “fully sufficient [90, 100];” “fully suffi-
disasters and emergencies occur; the more fully prepared the
cient” is a qualitative language description, and “90–100” is the
emergency plan, the stronger the disaster prevention and
corresponding quantitative score range. If an expert judged
mitigation capability. The factors related to disaster re- from experience that the situation of the X8 index to be
duction are divided into two categories, namely, emergency evaluated was fully sufficient, it would have been scored to be
rescue support factors and emergency rescue organization between 90 and 100 in the questionnaire.
factors. Copious professional rescue personnel and sufficient It must be pointed out that, to the best of the authors’
relief materials are the foundations of successful emergency knowledge, there has been no study on the upper limit values
response work. When natural disasters such as waterlogging of X1, X2, X3, X5, and X6. To better generalize the index
occur, the more complete the emergency rescue organiza- system, the upper limits of these indexes are “+∞” in Ta-
tion, the faster the information communication, and the
ble 1. When this index system was used to carry out case
higher the efficiency of emergency rescue.
studies, their upper limit values would have been reasonably
selected in combination with deep foundation pit engi-
neering practice, rather than being considered to be +∞.
2.1.2. Construction of Evaluation Index System. Taking fully
into account objectivity, systematism, availability, and in-
dependence, risk indicators that are representative of all risk 2.2. Risk Assessment Method of Waterlogging Risk in the Deep
factors identified in Section 2.1.1 were chosen in this study. Foundation Pits
An evaluation index system including 11 risk indicators was
constructed, as presented in Table 1. 2.2.1. Projection Pursuit Model. The PPM is a statistical
In Table 1, X1, X2, X3, X5, X6, X7, and X10 are method for processing and analysing high-dimensional data.
quantitative indicators, and their data was obtained by field The method was originally coined by Friedman and Tukey
4 Advances in Civil Engineering

Table 1: The index system of waterlogging risk in the deep foundation pit projects.
Indicator level Unit I II III IV
Intensity of rainfall
mm/24 h [0, 50) [50, 100) [100, 250) [250, +∞)
X1
Frequency of storms
times/year [0, 1) [1, 3) [3, 6) [6, +∞)
X2
Distance to urban drains
m [0, 50) [50, 200) [200, 500) [500, +∞)
X3
Surrounding environment No Few Moderate Many
—
X4 [90, 100] [75, 90) [60, 75) [0, 60)
Height of retaining wall
m [1.2, +∞) [0.6, 1.2) [0.3, 0.6) [0, 0.3)
X5
Number of pumps
/100 m2 [0.5, +∞) [0.2, 0.5) [0.1, 0.2) [0, 0.1)
X6
Proportion of older workers
% [0, 5) [5, 10) [10, 20) [20, 100]
X7
Preparation of emergency rescue plans Fully sufficient Sufficient Basically insufficient Insufficient
—
X8 [90, 100] [75, 90) [60, 75) [0, 60)
Efficiency of emergency rescue Fully efficient Efficient Basically efficient Inefficient
—
X9 [90, 100] [75, 90) [60, 75) [0, 60)
Proportion of relief workers
% [30, 100] [20, 30) [10, 20) [0, 10)
X10
Situation of emergency relief supplies Fully efficient Efficient Basically efficient Inefficient
—
X11 [90, 100] [75, 90) [60, 75) [0, 60)

from Stanford University in 1974 [32]. The basic idea is m

tantamount for projecting high-dimensional data into a low- Z(i) � 􏽘 aj xij∗ . (3)
dimensional space. Therefore, this model can effectively j�1

eliminate variable interference that is not related to the data

structure. Additionally, the objective weights of the indi- Optimizing a requires that the distribution character-
cators were computed directly from the characteristics of the istics of Z(i) are such that the projection points are locally as
sample data [33] in their application to system evaluation. dense as possible and overall spread out as much as possible.
The general steps of the PPM used in the system eval- Therefore, the best projection direction is determined by
uation are as follows. maximizing the product of the standard deviation and the
local density. The optimal projection function [36] is
(1) Standardization of Data 􏽳����������������
Let the indicator sequence be [xij ]m×n, where m is equal 􏽐ni�1 (Z(i) − E(z))2 (4)
Sz � ,
to the number of samples and n is the number of indicators. n−1
To eliminate the dimensional impact of the evaluation
indicators and ensure the generalizability of the modelling, where Sz is the standard deviation of Z(i) and E(z) is the
the extreme value normalization method was employed in average of Z(i).
this study to standardize the data. n m
The benefit-based indicators [34] are as follows: Dz � 􏽘 􏽘􏼐R − rij 􏼑u􏼐R − rij 􏼑, (5)
i�1 j�1
xij − min􏼐xj 􏼑
xij∗ � , (1) where Dz is the local density of Z(i), rij is the distance
max􏼐xj 􏼑 − min􏼐xj 􏼑
between samples, R is the window radius of the local density,
and the cost-based indicators [34] are as follows: and it is also the only parameter set in the PPM. Addi-
tionally, u(R − rij ) is the unit step function. When
max􏼐xj 􏼑 − xij R − rij ≥ 0, u(R − rij ) � 1; otherwise, u(R − rij ) � 0.
xij∗ � , (2)
max􏼐xj 􏼑 − min􏼐xj 􏼑 (3) Optimizing the Projection Index Function
where xij∗ indicates the value of the evaluation index after When the indicator sample set [xij ]m×n is determined, Q(a)
standardization, max(xj ) represents the maximum value of only changes with the projection direction a. If the projection
indicator j, and max(xj ) represents the minimum value of direction is the best projection direction a∗ , Q(a) must be at the
indicator j. maximum value. The low-dimensional data obtained in the best
projection direction a∗ can be best explained by the charac-
(2) Construction of Projection Indicator Function
teristic structure of the original data [37].
The core idea of the PPM is to project [xij∗ ]m×n to obtain Based on the preceding analysis, the method of finding
the projection value z(i) according to projection indicators the maximum value of the projection index function was
� (a1 , a2 , · · · , am ) [35]. The projection indicator function is employed to find the best projection direction.
Advances in Civil Engineering 5

The function to maximize the objective is [38] by the random sampling method. Equation (1) or (2)
􏼌􏼌 􏼌􏼌 is leveraged to normalize it to obtain [xij∗ ]p×n . To
max Q(a) � Sz 􏼌􏼌Dz 􏼌􏼌, (6)
ensure the accuracy and stability of subsequent in-
and the constraint is terpolation function models, hundreds of standard
m sample sets are extracted, so [xij∗ ]p×n is always high-
􏽘 a2j � 1, 0 ≤ aj ≤ 1. (7) dimensional data. Because the standard sample set
j�1 [xij ]p×n is generated according to the random
sampling method, the risk levels Y1 (i) are also
(4) Solving the Optimum Projection Direction known.
(2) Field research, theoretical computing, and ques-
Equations (6) and (7) are utilized to compute the optimal
tionnaires were used to obtain the evaluation indi-
projection vector. Generally, most scholars would use GA
cator set of the research object [xij ]m×n. According to
[13, 15, 16] and other algorithms for the solution. However,
the characteristics of each indicator, (1) or (2) is
the GA has drawbacks, including the computing result
adopted to standardize it and subsequently obtain
having certain dependence on the initial population selec-
[xij∗ ]m×n .
tion, its slow convergence speed, and its excessive parameter
settings [18, 19]. In contrast, PSO is characterized by ad- Step 2. Calculating the Weights and Projection
vantages including a fast convergence speed and fewer Values by the PPM and PSO
parameter settings [20–22, 39]; thus, PSO was employed in (1) [xij∗ ]p×n and [xij∗ ]m×n are combined into a computing
the present study to find the best projection vector a∗ . set [xij∗ ](m+p)×n . The projection index function Z(i) is
By substituting a∗ into (3), the projection vector Z∗ (i) constructed according to (3).
can be solved. (2) Equations (4) and (5) are used to find the maximum
objective function Q(a), and PSO is utilized to solve
2.2.2. Particle Swarm Optimization. PSO is a group-based the function Q(a).
random search algorithm that is designed by simulating the (3) When the PSO reaches the convergence condition,
predation behaviour of bird swarms. Its basic idea is to start the optimal projection direction a∗ is obtained. Z(i)
from a random solution, find the optimal solution through is then computed, where i � 1, 2, 3, · · · , m + p. The
iteration, and then evaluate and determine the optimal projected value Z1 (i) of the standard sample set
solution through fitness. [xij∗ ]p×n and the projected value Z2 (i) of the evalu-
In each iteration of PSO, a particle updates its own ation indicator set [xij∗ ]m×n are both included in Z(i).
position by tracking the individual optimal solution and the (4) Every element of a∗ is squared to obtain the objective
group optimal solution, thereby constantly adjusting its weight of each index [35].
position to approach the optimal position.
Let the population size of the particles be N; the velocity Step 3. Obtaining Evaluation Levels by the Inter-
update formula of the particle b is as follows: polation Algorithm
(1) According to the projection value of the standard
vb (t + 1) � ωvb (t) + c1 r􏼂qb (t) − ab (t)􏼃 + c2 r􏼂g(t) − ab (t)􏼃.
sample set Z1 (i) and its default risk level Y1 (i), the
(8) interpolation method is used to construct the math-
ematical model of risk assessment as follows [40]:
The velocity update formula is as follows:
ab (t + 1) � ab (t) + vb (t + 1), (9)
Y � f Z1 􏼁. (10)
where ω is the inertia weight factor, c1 and c2 are the learning
factors, r is a random number within the interval [0, 1], q is
(2) By introducing the projection value of the evaluation
the individual optimal value, and g is the global optimum
sample set Z2 (i) into the mathematical model
value.
Y � f(Z1 ), the waterlogging risk level of each
After reaching a certain condition in which the iteration
evaluation sample is computed.
terminates, the best projection vector a∗ is obtained by the
position of the particle swarm aggregation. From these steps, it is evident that this method is directly
motivated by data. Compared with functional mode eval-
uation methods, such as fuzzy comprehensive evaluation,
2.2.3. Risk Assessment Method Construction. The flow chart
the proposed method effectively avoids the difficulty of
of this evaluation method is illustrated in Figure 1.
constructing the functional relationship between the eval-
The specific steps of the construction of the risk as-
uation index set and evaluation level.
sessment method based on the PPM are as follows.
Step 1. Preparing Data
3. Results and Discussion
(1) According to Table 1, the random sampling method
is used to generate the standard sample set [xij ]p×n, 3.1. Engineering Background. The Chengdu Metro Line 11
where p is the number of standard samples obtained project involves a total investment of about 16.5 billion yuan.
6 Advances in Civil Engineering

Get
Random sampling in
standard

Data normalization
Construct projection indicators grading
sample

Preparing data
indicator function standard
data

Calculating weight and projection by PP method

Get
Local inspection and
evaluation
questionnaire
Calculate the optimum sample
survey
projection direction with PSO data

Projection value
Meet termination The risk level of
of standard

Obtaining evaluation level by the

conditions standard sample
sample data

interpolation algorithm
The projection
Piecewise linear
value of
interpolation
The optimum evaluation
Projection function
projection sample data
value
direction

Determine the risk level of

Determine the weights evaluation object

Figure 1: Flow chart of the waterlogging risk evaluation method based on the PPM.

It is about 22 km in length and comprises 22 station projects. elements. Xinchuan Science Park East Station is
This major project traverses the Chengdu Hi-tech Zone, situated in the suburban development zone, in which
Tianfu New District, and Shuangliu District. The landform there are fish ponds and a large number of gas
features are mostly plains, platforms, or a small number of pipelines.
low hills. The average annual rainfall in Chengdu is (3) Diaoyuzui Station is the largest station on Chengdu
879.3 mm, and the maximum rainfall in 24 h is 167.6 mm. Metro Line 11, with a total length of 340.40 m, a total
Waterlogging and droughts often occur in Chengdu. width of 21.5 m, and an excavation depth of
The deep foundation pit projects of Diaoyuzui Station, 16.9–27.3 m, all of which are the largest among the
Tianfu CBD North Station, and Xinchuan Science Park East three case study stations. Xinchuan Science Park East
Station were selected as a case study. The platforms of these Station is the smallest.
three stations are two-story islands. The deep foundation pits
were all constructed by the open-cut method, and there are
no adverse geological effects or special geotechnical soil 3.2. Sources of Data. Based on the hydrological and mete-
within their construction scopes. However, among all 22 orological characteristics of Chengdu and the 22 deep
station projects on Chengdu Metro Line 11, these three foundation pit projects of Chengdu Metro Line 11, the upper
projects have the following differences. limit values of X1, X2, X3, X5, and X6 were, respectively,
determined to be 500, 12, 3000, 2.4, and 1.
(1) The construction company used for Diaoyuzui The quantitative indicator scores of these three deep
Station and Xinchuan Science Park East Station, foundation pits were obtained by consulting the Chengdu
China Construction Third Bureau Group Co., Ltd., Water Resources Announcement, the project management
has rich experience in subway construction; how- documents of the Chengdu Metro Line 11 project, and field
ever, the construction company used for Tianfu CBD surveys. Their scores are presented in Table 2. The scores of
North Station is China Railway Investment and the quantitative indexes X4, X8, X9, and X11 were obtained
Construction Co., Ltd., which lacks experience in from a questionnaire survey of ten experts.
subway. Among the ten experts, three were from universities, two
(2) Diaoyuzui Station, which is located in the suburban were from construction companies, and five were from
countryside, has almost no urban drains in its sur- subway construction units. Seven experts had the title of
rounding area, which is characterized by environ- senior engineer or above, and the other three experts had
mental elements that are prone to waterlogging, such titles of associate senior engineer or associate professor. Six
as flood canals and high slopes. Tianfu CBD North experts were well aware of construction safety risks and have
Station is located in the CBD area that is under produced related research. Seven experts were familiar with
construction, where the municipal drains are well the construction of Chengdu Metro Line 11 and participated
developed and there are no adverse environmental in the construction of the project. SPSS 22 software was used
Advances in Civil Engineering 7

Table 2: Indicators’ scores of three evaluation objects.

Diaoyuzui Tianfu CBD North Xinchuan Science and Technology
Indicator Data sources
Station Station Park
Chengdu Water Resources
X1 167.6 167.6 167.6
Announcement
Chengdu Water Resources
X2 3.125 3.125 3.125
Announcement
X3 2600 150 50 Field research
X4 13.5 31 92 Field research and questionnaire survey
X5 0.6 1 0.4 Field research
X6 0.11 0.11 0.17 Field research
X7 8.61 9.09 11.72 Field research
X8 87.5 33 46.5 Field research and questionnaire survey
X9 73 70.5 80 Field research and questionnaire survey
X10 19.95 10.87 19.01 Field research
X11 73 72.5 79.5 Field research and questionnaire survey

to test the reliability of this questionnaire. The value of 840

Cronbach’s α was found to be 0.743 and therefore met the
reliability requirement of a questionnaire survey [41]; thus, 830
the results of this questionnaire survey were reliable. The 820
average of the scoring results of the 10 experts was the scores
of these quantitative indicators. 810
Fitness

800

3.3. Mathematical Risk Assessment Model with the Interpo- 790

lation Algorithm. According to Table 1 and Figure 1, 100
780
standard evaluation objects in each risk level were generated
by the random sampling method [39]. Thus, there were 400 770
standard evaluation objects obtained for the establishment
760
of a mathematical model of risk evaluation. The data of these
0 200 400 600 800 1000
400 standard evaluation objects (the standard sample data in
Number of iterations
Figure 1) and the data of 3 units to be assessed (the eval-
uation sample data in Figure 1) were substituted into a self- Figure 2: The convergence curve of the PPM optimized by PSO.
programmed program based on MATLAB R2016a software.
Referring to previous research results [42–45], the swarm
size was 200, the personal learning coefficient and global After calculation, the best projection direction a∗ was
learning coefficient were both 2, the inertia weights de- found to be (0.3642, 0.3784, 0.3225, 0.3207, 0.1765, 0.1887,
creased linearly from 0.9 to 0.4, the minimum acceptance 0.3340, 0.3481, 0.31860, 0.1004, 0.3258). The scatter diagram
accuracy was 0.00001, and the maximum number of itera- of the projection values Z1 (i) of 400 standard evaluation
tions was 1000. Although the minimum accuracy require- objects and the corresponding risk level Y1 (i) is presented in
ment was met when the iteration number reached about 200 Figure 3.
in the case analysis, the population number and the max- As exhibited in Figure 3, the scatter diagram of the best
imum iteration number were set to be relatively large in this projection values and risk levels was a ladder-type and was
study to ensure that the model could calculate more complex characterized by an increasing curve; the larger the pro-
problems. The convergence curve of the 1000 iterations is jection value, the greater the waterlogging risk. The pro-
presented in Figure 2. jection values were aggregated rather than being
Following the optimization calculation process of PSO, continuously distributed. This graphic feature was deter-
the error between the 194th iteration and 195th iteration was mined by the basic idea of the PPM. Locally, the optimized
greater than the minimum acceptance precision (0.00001), projection points should be clustered as much as possible.
and the error between the 195th iteration and 196th iteration On the whole, they should be spread as much as possible.
was less than that of the minimum acceptance precision This clustering phenomenon presented in Figure 3 was fairly
stage. After that, the errors of the calculation results were all consistent with the results of previous classical literature
less than 0.00001. Based on the calculation termination [46], and Figure 3 further demonstrates that the PPM uti-
conditions of the algorithm, the calculation was arrested at lized in this study had been optimized.
the 1000th iteration with a very small error. These findings The maximum projection value in the low risk (I) level
indicate that the PSO algorithm found the best projection was 0.3818. The maximum projection value in the moderate
vector at the 196th iteration, which is illustrated by both risk (II) level was 0.8562, and the minimum projection value
Figure 2 and Table 3. was 0.6079. The maximum projection value in the high risk
8 Advances in Civil Engineering

Table 3: Precision level and calculation termination in the 1000th iteration.

Iteration (n) Fitness (n − 1) Fitness (n) Fitness (n) − Fitness (n − 1) Result
194 834.956821 834.956821 0 < 0.00001 Continue
195 834.956821 834.956841 0.0000197 > 0.0001 Continue
196 834.956841 834.956841 0 < 0.0001 Continue
1000 834.956842 834.956842 0 < 0.0001 Stop

perspectives of environmental science and civil engineering;

4 if this research had been carried out from the perspectives of
other disciplines, the index system and calculation results of
the weights and risk levels would have been different. Due to
3 the lack of academic research on the waterlogging risk as-
sessment of deep foundation pit engineering, the calculation
Risk level

results are primarily compared and discussed with the actual

disaster situation of Chengdu Metro Line 11 in the 2018
2
flood season. In addition, while the PSO was successfully
employed to optimize the PPM in the present case study,
many other optimization calculation methods could be
1 applied for the optimization of the PPM. Depending on the
results of the case analysis, this section also presents detailed
0.0 0.5 1.0 1.5 2.0 2.5 3.0 measures suggested for addressing the waterlogging risk in
The best projection value
deep foundation pit engineering.

Figure 3: Scatter diagram of the standard evaluation object pro-

jection values and risk levels. 3.4.1. Weight Analysis. After squaring each element in the
calculated optimal projection direction [35], namely (0.3642,
(III) level was 1.4125, and the minimum projection value was 0.3784, 0.3225, 0.3207, 0.1765, 0.1887, 0.3340, 0.3481,
1.1010. The minimum projection value in the extreme risk 0.31860, 0.1004, 0.3258), the objective weights of the 11
(IV) level was 1.9991. Using the piecewise linear interpo- indicators were calculated and are presented in Table 4.
lation method, a mathematical model for the waterlogging From Table 4, it is evident that X2 (frequency of storms)
risk assessment of deep foundation pit engineering was and X1 (intensity of rainfall) had the greatest impacts on the
established: risk of waterlogging in deep foundation pit engineering.
These results were similar to the calculation results of a
⎪
⎧ 1, z∗ ≤ 0.3818,
⎪
⎪ previous study [47]. From the perspective of disaster science,
⎪
⎪
⎪
⎪ X1 and X2 were found to be the greatest indicators, which
⎪ ∗
⎪ 1 + z − 0.3818 , 0.3818 < z∗ < 0.6079,
⎪
⎪ demonstrates that disaster-causing factors have the greatest
⎪
⎪ 0.6079 − 0.3818
⎪
⎪ impacts on waterlogging risk.
⎪
⎪ The weight of X8 ranked third and was also the largest
⎪
⎪
⎪
⎪
⎪
⎪ 2, 0.6079 ≤ z∗ ≤ 0.8562, among the indicators related to disaster prevention and
⎪
⎪ mitigation. This further demonstrates that emergency rescue
⎪
⎪
⎪
⎪ plans form the basis of emergency management. The weight
⎨ z∗ − 0.8562
y(i) � ⎪ 2 + , 0.8562 < z∗ < 1.1010, of X7 (proportion of older workers) ranked fourth and
⎪
⎪ 1.1010 − 0.8562
⎪
⎪ therefore also has a great influence on the waterlogging risk.
⎪
⎪
⎪
⎪ It is worth noting that the superior weight of X7 might be
⎪
⎪ 3, 1.1010 ≤ z∗ ≤ 1.4125,
⎪
⎪ closely related to the aging phenomenon of Chinese con-
⎪
⎪
⎪
⎪ struction workers in the past decade.
⎪
⎪ z∗ − 1.4125
⎪
⎪ The weight analysis results were found to be in agree-
⎪
⎪ 3+ , 1.4125 < z∗ < 1.9991,
⎪
⎪ 1.9991 − 1.4125 ment with the actual disaster situation of the deep foun-
⎪
⎪
⎪
⎪ dation pit project of Diaoyuzui Station during the flood
⎩
4, z∗ ≥ 1.9991. season of 2018. At the beginning of July 2018, a series of
(11) torrential rainstorms in Chengdu led to the inflow of water
in the deep foundation pit of Diaoyuzui Station. When the
waterlogging disaster occurred, the project managers pri-
3.4. Analysis and Discussion. This section analyses and oritized saving lives, so the workers at the construction site
discusses the calculation results of the weights and risk levels and in the living area, especially the older workers, were first
in detail. However, the waterlogging risk evaluation index evacuated, resulting in a delay in the rescue of the site,
system presented in this work was constructed from the thereby enlarging the disaster losses.
Advances in Civil Engineering 9

Table 4: Weights and rankings of all risk indicators.

Indicator X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11
Weight 0.1327 0.1432 0.1040 0.1029 0.0312 0.0356 0.1116 0.1212 0.1015 0.0101 0.1061
Ranking 2 1 6 7 10 9 4 3 8 11 5

Based on these results, some advice is provided for decrease the level of waterlogging risk as soon as
project managers regarding the carrying out of the risk possible.
management and decision-making related to waterlogging (2) In the design and planning of the construction or-
disasters in deep foundation pit projects; focus should be ganization of deep foundation pit projects, the local
placed on X2 (frequency of storms), X1 (intensity of rain- hydrometeorological data, especially the data of
fall), X8 (preparation of emergency rescue plans), and X7 regional annual rainfall frequency and the largest
(proportion of older workers), while the investment of re- rainfall intensity in history, should be fully investi-
sources in factors X5, X6, and X10 should be of secondary gated. During construction, project managers should
concern. consider future climate changes and adjust the
measures for waterlogging risk according to the
3.4.2. Risk Level Analysis. The best projected values for the changes in weather.
deep foundation pit projects of Diaoyuzui Station, Tianfu (3) The managers of deep foundation pit projects should
CBD North Station, and Xinchuan Science Park East Station attach more importance to the preparation of
were found to be 1.4524, 1.3275, and 1.0644, respectively. By emergency rescue plans. Before the flood season,
substituting these values into (11), the respective risk levels they should conduct emergency rescue drills for
were determined to be 3.0681, 3.0000, and 2.8505. The waterlogging, strengthen the construction of rescue
waterlogging risk level of the Diaoyuzui Station was between organizations, and improve the efficiency of emer-
the high risk (III) and the extreme risk (IV) levels. The gency rescue.
waterlogging risk in Tianfu CBD station was at the high risk (4) The proportion of older workers should be reduced
(III) level and that of Xinchuan Science Park East Station was to mitigate the vulnerability of the local population.
between the moderate risk (II) and high risk (III) levels. However, in China, the aging of construction
Thus, the waterlogging risks in these three deep foundation workers has become increasingly more severe in the
pits from greatest to least were found to be as follows: past ten years, and it is a difficult task to reduce the
Diaoyuzui Station > Tianfu CBD North Station > Xinchuan proportion of older employees. Therefore, as an al-
Science Park East Station. ternative measure, increasing emergency rescue
During the 2018 flood season in Chengdu, Diaoyuzui training is recommended.
Station was the most severely damaged of all the 22 station
projects. Tianfu CBD North Station was the second-most 4. Conclusions
damaged, but Xinchuan Science Park East Station was hardly
affected by this flood disaster. The actual disaster situations The intention of this study was to conduct an objective and
of the three deep foundation pit projects from greatest to effective evaluation of waterlogging risk in deep foundation
least were therefore as follows: Diaoyuzui Station > Tianfu pits based on the projection pursuit method. According to
CBD North Station > Xinchuan Science Park East Station the process of water circulation and the characteristics of
Project. This ranking obtained by the on-site investigation deep foundation pits, an evaluation index system with 11
was consistent with the waterlogging risk results calculated indicators was constructed for the first time. Then, a
by the method proposed in this study. The consistency combination of the projection pursuit method, particle
proves that the proposed method is both objective and swarm optimization, and the interpolation algorithm was
effective. used to construct a waterlogging risk assessment method. In
this method, the projection pursuit method is used to handle
the high-dimensional data of waterlogging risk assessment,
3.4.3. Measures to Address Waterlogging Risk in Deep
and the optimal projection vector is solved by particle swarm
Foundation Pit Projects
optimization, which is characterized by relatively rapid and
(1) To better ensure construction safety, the deep appropriate convergence. The mathematical function be-
foundation pit project of Diaoyuzui Station should tween the best projection values and the risk levels was
deal with risks in accordance with the extreme risk constructed by an interpolation algorithm. Finally, three
(IV) level. Project managers should immediately typical deep foundation pits of Chengdu Metro Line 11 were
suspend construction tasks and take steps to reduce analysed, and some risk response strategies were provided
the risk level. The construction operations of Tianfu based on the results of the case analysis. The results of the
CBD North Station and Xinchuan Science Park East case study revealed that the frequency of storms, intensity of
Station do not require termination, but their project rainfall, preparation of emergency rescue plans, and pro-
management staff should immediately formulate portion of older workers have the greatest impacts on the
further measures to address waterlogging risk and waterlogging risk in deep foundation pits. These four factors
10 Advances in Civil Engineering

were also the key factors in the waterlogging disaster of [8] R. Mikaeil, S. Shaffiee Haghshenas, Y. Shirvand, M. Valizadeh
Diaoyuzui Station during the 2018 flood season, and project Hasanluy, and V. Roshanaei, “Risk assessment of geological
management personnel should place focus on them in future hazards in a tunneling project using harmony search algo-
waterlogging risk management and decision-making. The rithm (case study: Ardabil-Mianeh Railway tunnel),” Civil
waterlogging risk ranking results of the three typical deep Engineering Journal, vol. 2, no. 10, pp. 546–554, 2016.
[9] D. T. Bui, N. D. Hoang, T. D. Pham et al., “A new intelligence
foundation pits calculated by the proposed method were
approach based on GIS-based multivariate adaptive regression
basically congruent with the disaster situation in the 2018 splines and metaheuristic optimization for predicting flash
flood season. Consequently, it can be concluded that the flood susceptible areas at high-frequency tropical typhoon
proposed method based on the projection pursuit method is area,” Journal of Hydrology, vol. 575, pp. 314–326, 2019.
both objective and effective. Future research will concentrate [10] R. Mikaeil, M. Beigmohammadi, E. Bakhtavar et al., “As-
on the establishment of a unified waterlogging risk evalu- sessment of risks of tunneling project in Iran using artificial
ation index system and the use of more soft computing bee colony algorithm,” SN Applied Sciences, vol. 1, no. 12,
methods for waterlogging risk evaluation. Article ID 1711, 2019.
[11] P. Wei, F. Qiang, L. Dong et al., “Assessing agricultural
drought vulnerability in the Sanjiang Plain based on an
Data Availability improved projection pursuit model,” Natural Hazards,
vol. 82, no. 1, pp. 683–701, 2016.
The MATLAB programs and case analysis data used to
[12] Y. Sen and L. H. Wei, “An integrated model of water resources
support the findings of this study are available from the optimization allocation based on projection pursuit model-
corresponding author upon request. grey wolf optimization method in a transboundary river
basin,” Journal of Hydrology, vol. 559, pp. 156–165, 2018.
Conflicts of Interest [13] Y. Liang, D. Guo, Z. Huang, and X. Jiang, “Prediction model
for coal-gas outburst using the genetic projection pursuit
The authors declare that there are no conflicts of interest method,” International Journal of Oil, Gas and Coal Tech-
regarding the publication of this paper. nology, vol. 16, no. 3, pp. 271–282, 2017.
[14] D. Liu, G. D. Zhang, and H. Li, “Projection pursuit evaluation
model of a regional surface water environment based on an
Acknowledgments Ameliorative Moth-Flame Optimization algorithm,” Ecolog-
ical Indicators, vol. 107, no. 1, Article ID 105674, 2019.
This study was supported by the National Key R&D Program [15] Y. S. Jiang, D. Zhao, D. D. Wang et al., “Sustainable per-
of China (2018YFC0704301) and the Science and Tech- formance of buildings through modular prefabrication in the
nology Project of Wuhan Urban and Rural Construction construction phase: a comparative study,” Sustainability,
Bureau, China (201943). vol. 11, no. 20, Article ID 5658, 2019.
[16] S. Li, C. Jiang, and A. Li, “Research on Chinese well-known
e-commerce enterprises’ innovation ability based on real
References comment,” International Journal of Computing Science and
[1] S. Tei, T. Morozumi, S. Nagai et al., “An extreme flood caused Mathematics, vol. 11, no. 1, pp. 54–62, 2020.
by a heavy snowfall over the Indigirka river basin in [17] W. J. Zhang, F. Wei, R. Zhou et al., “Risk assessment of cotton
Northeastern Siberia,” Hydrological Processes, vol. 34, no. 3, textile enterprise working environment based on projection
pp. 522–537, 2020. pursuit model,” China Safety Science Journal, vol. 28, no. 4,
[2] S. Liu, M. N. Lin, and C. L. Li, “Analysis of the effects of the pp. 103–108, 2018.
river network structure and urbanization on waterlogging in [18] A. Kukker and R. Sharma, “Genetic algorithm-optimized
high-density urban areas-a case study of the Pudong new area fuzzy lyapunov reinforcement learning for nonlinear sys-
in Shanghai,” International Journal of Environmental Research tems,” Arabian Journal for Science and Engineering, vol. 45,
and Public Health, vol. 16, no. 18, Article ID 3306, 2019. no. 3, pp. 1629–1638, 2020.
[3] H. Wu, “Research and discussion on urban waterlogging in [19] R. Bendaoud, H. Amiry, M. Benhmida et al., “New method for
Wuhan,” Water Supply and Sewerage, vol. 53, no. 1, extracting physical parameters of PV generators combining
pp. 117–119, 2017. an implemented genetic algorithm and the simulated
[4] J. Hou and Y. Du, “Spatial simulation of rainstorm water- annealing algorithm,” Solar Energy, vol. 194, pp. 239–247,
logging based on a water accumulation diffusion algorithm,” 2019.
Geomatics, Natural Hazards and Risk, vol. 11, no. 1, pp. 71–87, [20] R. Mikaeil, S. Shaffiee Haghshenas, and Z. Sedaghati, “Geo-
2020. technical risk evaluation of tunneling projects using opti-
[5] J. X. Jia, X. Y. Wang, N. A. M. Hersi et al., “Flood-risk zoning mization techniques (case study: the second part of
based on analytic hierarchy process and fuzzy variable set Emamzade Hashem tunnel),” Natural Hazards, vol. 97, no. 3,
theory,” Natural Hazards Review, vol. 20, no. 3, Article ID pp. 1099–1113, 2019.
04019006, 2019. [21] A. R. Dormishi, M. Ataei, R. K. Kakaie et al., “Performance
[6] Z. N. Wu, Y. X. Shen, and H. L. Wang, “Assessing urban areas’ evaluation of gang saw using hybrid ANFIS-DE and hybrid
vulnerability to flood disaster based on text data: a case study ANFIS-PSO algorithms,” Journal of Mining and Environment,
in Zhengzhou city,” Sustainability, vol. 11, no. 17, Article ID vol. 10, no. 2, pp. 543–557, 2019.
4548, 2019. [22] M. Hasanipanah, R. Naderi, J. Kashir, S. A. Noorani, and
[7] H. Y. Yu, C. Liang, P. Li et al., “Evaluation of waterlogging risk A. Zeynali Aaq Qaleh, “Prediction of blast-produced ground
in an urban subway station,” Advances in Civil Engineering, vibration using particle swarm optimization,” Engineering
Article ID 5393171, 2019. with Computers, vol. 33, no. 2, pp. 173–179, 2017.
Advances in Civil Engineering 11

[23] P. Ronco, V. Gallina, S. Torresan et al., “The KULTURisk Problems in Engineering, vol. 2017, Article ID 2902506,
regional risk assessment methodology for water-related 13 pages, 2017.
natural hazards–part 1: physical-environmental assessment,” [39] C. Qi, A. Fourie, Q. Chen et al., “Neural network and particle
Hydrology and Earth System Sciences, vol. 18, no. 12, swarm optimization for predicting the unconfined com-
pp. 5399–5414, 2014. pressive strength of cemented paste backfill,” Construction
[24] P. Cui and D. Li, “Measuring the disaster resilience of an and Building Materials, vol. 159, pp. 473–478, 2018.
urban community using ANP-FCE method from the per- [40] X. H. Yang, Z. F. Yang, Z. Y. Shen et al., “Interpolation model
spective of capitals,” Social Science Quarterly, vol. 100, no. 6, for flood disaster assessment based on projection pursuit,”
pp. 2059–2077, 2019. Disaster Science, vol. 4, pp. 3–8, 2004.
[25] Y. Zhou, N. Li, W. Wu, and J. Wu, “Assessment of provincial [41] J. E. Ware, M. Kosinski, and S. D. Keller, “A 12-item short-
social vulnerability to natural disasters in China,” Natural form health survey,” Medical Care, vol. 34, no. 3, pp. 220–233,
Hazards, vol. 71, no. 3, pp. 2165–2186, 2014. 1996.
[26] Y. Wei and O. Linet, “A dynamic logistics coordination model [42] J. J. Liang, A. K. Qin, P. N. Suganthan, and S. Baskar,
for evacuation and support in disaster response activities,” “Comprehensive learning particle swarm optimizer for global
European Journal of Operational Research, vol. 179, no. 3, optimization of multimodal functions,” IEEE Transactions on
pp. 1177–1193, 2007. Evolutionary Computation, vol. 10, no. 3, pp. 281–295, 2006.
[27] M.-S. Chang, Y.-L. Tseng, and J.-W. Chen, “A scenario [43] N. Lynn and P. N. Suganthan, “Heterogeneous comprehen-
planning approach for the flood emergency logistics prepa- sive learning particle swarm optimization with enhanced
ration problem under uncertainty,” Transportation Research exploration and exploitation,” Swarm and Evolutionary
Part E: Logistics and Transportation Review, vol. 43, no. 6, Computation, vol. 24, pp. 11–24, 2015.
pp. 737–754, 2007. [44] Z. H. Zhan, J. Zhang, Y. Li, and H. S.-H. Chung, “Adaptive
[28] E. L. Guo, J. Q. Zhang, Y. F. Wang et al., “Dynamic risk particle swarm optimization,” IEEE Transactions on Systems
assessment of waterlogging disaster for maize based on CE- Man and Cybernetics Part B-Cybernetics, vol. 39, no. 6,
RES-Maize model in Midwest of Jilin province, China,” pp. 1362–1381, 2009.
Natural Hazards, vol. 83, no. 3, pp. 1747–1761, 2016. [45] A. Berro, S. Larabi Marie-Sainte, and A. Ruiz-Gazen, “Genetic
[29] X. Liu and H. Chen, “Integrated assessment of ecological risk algorithms and particle swarm optimization for exploratory
for multi-hazards in Guangdong province in southeastern projection pursuit,” Annals of Mathematics and Artificial
China,” Geomatics, Natural Hazards and Risk, vol. 10, no. 1, Intelligence, vol. 60, no. 1-2, pp. 153–178, 2010.
pp. 2069–2093, 2019. [46] R. J. Bolton and W. J. Krzanowski, “Projection pursuit
[30] Y. L. Liang, Y. L. Wang, Y. J. Zhao et al., “Analysis and clustering for exploratory data analysis,” Journal of Compu-
projection of flood hazards over China,” Water, vol. 11, no. 5, tational and Graphical Statistics, vol. 12, no. 1, pp. 121–142,
Article ID 1022, 2019. 2003.
[31] M. Ma, H. Wang, P. Jia et al., “Investigation of inducements [47] A. I. Gorshkov, H. M. Hassan, and O. V. Novikova, “Seis-
and defenses of flash floods and urban waterlogging in mogenic nodes (M ≥ 5.0) in northeast Egypt and implications
Fuzhou, China, from 1950 to 2010,” Natural Hazards, vol. 91, for seismic hazard assessment,” Pure and Applied Geophysics,
no. 2, pp. 803–818, 2018. vol. 176, no. 2, pp. 593–610, 2019.
[32] D. Liu, C. Liu, Q. Fu et al., “Projection pursuit evaluation
model of regional surface water environment based on im-
proved chicken swarm optimization algorithm,” Water Re-
sources Management, vol. 32, no. 4, pp. 1325–1342, 2018.
[33] J. Zhao, J. Jin, Q. Guo, L. Liu, Y. Chen, and M. Pan, “Dynamic
risk assessment model for flood disaster on a projection
pursuit cluster and its application,” Stochastic Environmental
Research and Risk Assessment, vol. 28, no. 8, pp. 2175–2183,
2014.
[34] S. Wang, X. Zhang, Z. Yang, J. Ding, and Z. Shen, “Projection
pursuit cluster model based on genetic algorithm and its
application in Karstic water pollution evaluation,” Interna-
tional Journal of Environment and Pollution, vol. 28, no. 3/4,
pp. 253–260, 2006.
[35] D. Liu, J. P. Feng, H. Li et al., “Spatiotemporal variation
analysis of regional flood disaster resilience capability using an
improved projection pursuit model based on the wind-driven
optimization algorithm,” Journal of Cleaner Production,
vol. 241, Article ID 118406, 2019.
[36] X. Huang and H. Bai, “Risk prediction of rural public security
environmental carrying capacity based on the risk entropy,”
Natural Hazards, vol. 90, no. 1, pp. 157–171, 2018.
[37] Z. Lan and M. Huang, “Safety assessment for seawall based on
constrained maximum entropy projection pursuit model,”
Natural Hazards, vol. 91, no. 3, pp. 1165–1178, 2018.
[38] J. H. Yuan and C. B. Li, “Intuitionistic trapezoidal fuzzy group
decision-making based on prospect choquet integral operator
and grey projection pursuit dynamic cluster,” Mathematical