water

Article
Seepage Comprehensive Evaluation of Concrete Dam
Based on Grey Cluster Analysis
Junjie Li 1 , Xudong Chen 1, *, Chongshi Gu 2,3 and Zhongyan Huo 4
1 College of Water Conservancy & Environmental Engineering, Zhengzhou University, Zhengzhou 450001,
China
2 State Key Laboratory of Hydro-Water Resources and Hydraulic Engineering, Hohai University,
Nanjing 210098, China
3 College of Water Conservancy and Hydropower Engineering, Hohai University, Nanjing 210098, China
4 School of Port and Transportation Engineering, Zhejiang Ocean University, Zhoushan 316000, China
* Correspondence: chenxudong@zzu.edu.cn; Tel.: +86-183-3630-2520




Received: 17 June 2019; Accepted: 17 July 2019; Published: 19 July 2019


Abstract: Most concrete dams have seepage problems to some degree, so it is a common strategy to
maintain ongoing monitoring and take timely repair measures. In order to grasp the real operation
state of dam seepage, it is vital to analyze the measured data of each monitoring indicator and
establish an appropriate prediction equation. However, dam seepage states under the load and
environmental influences are very complicated, involving various monitoring indicators and multiple
monitoring points of each indicator. For the purpose of maintaining the temporal continuity and
spatial correlation of monitoring objects, this paper used a multi-indicator grey clustering analysis
model to explore the grey correlation among various indicators, and realized a comprehensive
evaluation of a dam seepage state by computation of the clustering coefficient. The case study
shows that the proposed method can be successfully applied to the health monitoring of concrete
dam seepage.

Keywords: dam seepage; comprehensive indicator system; seepage monitoring model; grey
clustering analysis

1. Introduction
After a dam is completed and impounded, once the seepage factors (seepage head, seepage
gradient, etc.) exceed the allowable value, the dam will suffer from seepage damage. The related
statistic shows that dam failures caused by seepage problems account for 30% to 40% of the total
number of failures, second only to flood overtopping [1]. Therefore, it is extremely important to carry
out comprehensive monitoring of dam seepage, which is the most effective way to monitor operating
conditions, find unsafe factors, and prevent problems before accidents occur [2]. However, dam
seepage states are affected by many external factors, such as water level, rainfall, temperature of dam
concrete, and time effects [3], which makes comprehensive seepage evaluation very complicated. It is
essential to find the variation rules of load set with dam seepage, as well as exploring the relationship
between them. The reasonable analysis method is to select the factors and expressions of the seepage
monitoring variables by deterministic function [4] and statistical correlation [5] according to the
basic principle of seepage flow, and then calculate the coefficients of the model with mathematical
statistics [6] based on monitoring data. By using mathematical expressions between independent
variables and dependent variables, the monitoring data can be fitted and the variation law of seepage
in the near future can be predicted.

Water 2019, 11, 1499; doi:10.3390/w11071499 www.mdpi.com/journal/water

Water 2019, 11, 1499 2 of 16

To date, studies [7–9] on the evaluation of dam seepage state have been mainly aimed at a
single evaluation indicator, which can only reflect partial information of the dam’s seepage state
and will therefore lead to inaccuracy and incompleteness of the evaluation results. In practical
engineering, the seepage state of concrete dams should be reflected by several indicators (dam seepage
pressure, dam leakage, and seepage around dam) [10], and multiple monitoring points of each
indicator distributed at different elevations in different dam sections. It is difficult to evaluate dam
seepage states directly according to the information of the monitoring points, so it is impossible
to know whether there is an abnormal phenomenon of seepage during the operation of the dam.
In order to realize the comprehensive evaluation of dam seepage safety, it is necessary to fuse the
information of multiple indicators with multiple monitoring points and explore the correlations among
various indicators, which is the main purpose and novelty of this article. There are many uncertain
problems [11] in the process of multi-information fusion, including incomplete information, incomplete
data, incommensurability of multiple indicators, etc. The commonly used ladder fusion methods
mainly include the following: set pair analysis theory [12], neural network [13], risk reliability combined
theory [14], matter element extension model [9], cluster fusion diagnosis model [15], cloud model [16],
defuzzification method [17], etc. However, most concrete dams built many years ago have limitations
of monitoring conditions and the randomness of manual monitoring, resulting in the problem of short
monitoring data sequences and inaccurate data. As a method of dealing with uncertain systems, small
samples, and poor information, grey clustering analysis theory can extract valuable information by
generating and mining [18], which is relatively suitable for the comprehensive evaluation of dam
seepage safety. Combining this theory with a seepage monitoring model, a multi-indicator grey
clustering comprehensive analysis model of dam seepage has been constructed, which integrates
the monitoring data of each indicator according to defined categories, and generates a whitenization
weight function value [19] and comprehensive clustering coefficient. In this process, the model outputs
the grey correlation degree [20] between different indicators and analyzes the change tendency of the
seepage state, and then divides the seepage state into different grey classes according to the fluctuations
of the monitoring data. This can give the staff a degree of judgment on the seepage state and provide
beneficial reference values for dam maintenance, so that we can visually find the parts with abnormal
seepage flow to prevent the occurrence of dangerous accidents.
This article is organized as follows. Section 2 introduces the seepage monitoring model of concrete
dams, which includes the establishment of a comprehensive indicator system of dam seepage state and
a statistical model, as well as the standard of grading. The grey clustering analysis theory is described
in Section 3.1, and a multi-indicator grey clustering comprehensive analysis model is established in
Section 3.2. The application of engineering cases is introduced in Section 4. Concluding remarks
complete the paper in Section 5.

2. Seepage Monitoring Model of Concrete Dams

2.1. The Comprehensive Indicator System of Dam Seepage State

The evaluation indicators of dam seepage state constitute a multi-level and multi-objective
comprehensive indicator system, which is mainly drawn from the following three aspects: (1) Dam
seepage pressure is composed of uplift pressure of the dam foundation and seepage pressure of the
dam body. Both are caused by seepage water and have certain influences on dam stability, deformation,
and stress; (2) Dam leakage causes the water to take the fine particles out of the dam body and form a
seepage passage, which endangers the stability of the dam; (3) Upstream impoundment can not only
seep through the dam body and foundation, but also seep downward around the bank slopes at both
ends of the dam. Considering the temporal continuity and spatial correlation [21] of the monitoring
points, each indicator is jointly monitored by several monitoring points for a long time series. With
reference to the monitoring data and the dam seepage monitoring research results [10,22,23], the
comprehensive indicator system of concrete dam seepage state is shown in Figure 1.
Water 2019, 11, 1499 3 of 16
Water 2019, 11, x FOR PEER REVIEW 3 of 16

Figure 1. Comprehensive
Figure 1. Comprehensive evaluation
evaluation indicator
indicator system
system of
of concrete
concrete dam
dam seepage.
seepage.

2.2. Seepage Monitoring Model

2.2. Seepage Monitoring Model
According to dam safety monitoring theory [24,25], concrete dam seepage is composed of four
According to dam safety monitoring theory [24,25], concrete dam seepage is composed of four
components: water level component yH , rainfall component yP , temperature component yT , and time
components: water level component 𝑦 , rainfall component 𝑦 , temperature component 𝑦 , and
effect component yθ . It can be expressed as follows:
time effect component 𝑦 . It can be expressed as follows:
𝑦 y=H𝑦+
yt = +
y 𝑦++yT𝑦𝑇++y𝑦θ𝜃 =
𝐻 P 𝑃
f (H,𝑃,P,𝑇,T,𝜃)θ)
= 𝑓(𝐻, (1)
(1)
The four
The four components
components are
are described
described below.
below.

2.2.1. Water
2.2.1. Water Level Component y𝑦H
Level Component
Through analysis
Through analysis of
of test
test data
data and
and seepage
seepage theory
theory [4]
[4] to
to deduce
deduce the
the expression
expression form
form of
of the
the water
water
pressure effect, there is a certain hysteresis and linear dependence between the change of
pressure effect, there is a certain hysteresis and linear dependence between the change of upstream upstream
water level
water level with
with uplift
uplift pressure
pressure of of dam
dam foundation
foundation and
and seepage
seepage around
around dam,
dam, so
so the
the water
water level
level
component is
component is usually
usually expressed
expressed as as follows:
follows:

5
𝑦 = X 𝑎 (ℎ − ℎ ) (2)
yH = ai (hi − h0i ) (2)
i=1
where 𝑎 (i = 1–5) is the regression coefficient of the water level component; ℎ (i = 1–5) is the
where
upstreamai (i = 1–5) level
water is the of
regression coefficient
the monitoring of the
day, water
1 day levelthe
before component;
monitoring = 1–5)
hi (iday, is the upstream
average upstream
water
water level
levelofofthe
2 tomonitoring
4 days beforeday,the
1 day before the
monitoring monitoring
day, 5 to 15 days day,before
average upstream
the monitoring water level
day, andof162
to
to 430days
daysbefore
beforethethemonitoring
monitoringday, day.5 toℎ 15(idays
= 1–5)before
is the the monitoring
average upstream day,water
and 16 to 30
level days before
corresponding
the monitoring
to each of the above h0i (i = 1–5)
day. periods on theis initial
the average upstream
monitoring day. water level corresponding to each of the
aboveTheperiods on the initial
permeability monitoring
coefficient day. is small, the aggregate gradation is different from the
of concrete
Thematerial,
particle permeability
and coefficient of concrete
the permeability is small,
equation the aggregate
is complex. gradation
Therefore, besidesis different from the
the first power of
particle material, and the permeability equation is complex.
upstream water depth, the water level component may also be related to the higher Therefore, besides the first power of
of
upstream
upstreamwater waterdepth,
depth,the sowater level component
it is taken to the fourth may
poweralso be
to related
simplifytosimulation
the higher power
of the of upstream
influence of
water
upstreamdepth, so it
water is taken
level on the toseepage
the fourth poweroftothe
pressure simplify
dam body,simulation of the influence
with reference of upstream
to the literature [26].
water levellevel
The water on the seepage pressure
component of the dam
of the seepage body,ofwith
pressure reference
the dam body tois the literature
often expressed [26].as The water
follows:
level component of the seepage pressure of the dam body is often expressed as follows:
𝑦 = 𝑎 ℎ −ℎ (3)
X4  
i i
yH = ai hu1 − hu0 (3)
where 𝑎 (i = 1–4) is the regression coefficient i=1 of the water level component; ℎ is the upstream
water depth of the monitoring day; and ℎ is the upstream water depth of the initial monitoring
day. ai (i = 1–4) is the regression coefficient of the water level component; hu1 is the upstream water
where
depthBased
of theon monitoring
the derivationday; of
and hu0 is the
seepage upstream
theory water
[26], the depth
leakage ofof the
the initial
bank monitoring
slope and riverday. bed dam
Based on the derivation of seepage theory [26], the leakage of
section is obtained. The dam leakage is related to the first and second power of upstreamthe bank slope and river bed
water dam
depth.
section is obtained.
At the same time, the The dam leakage
hysteretic effectisof
related to the
reservoir first level
water and second
on the power of upstream
dam leakage water depth.
is considered. The
water level component of dam leakage is often expressed as follows:
Water 2019, 11, 1499 4 of 16

At the same time, the hysteretic effect of reservoir water level on the dam leakage is considered. The
water level component of dam leakage is often expressed as follows:

2
X   X7  
yH = ai hu1 i − hu0 i + ai hu(i−2) − hu0 (4)
i=1 i=3

where ai (i = 1–7) are the regression coefficients of the water level component; hu(i−2) are the upstream
water level of the monitoring day, 1 day before the monitoring day, average upstream water level of 2
to 4 days before the monitoring day, 5 to 15 days before the monitoring day, and 16 to 30 days before
the monitoring day.

2.2.2. Rainfall Component yP

The seepage pressure of the dam foundations and both sides of dam is affected by groundwater,
which is mainly caused by rainfall in addition to the reservoir water level. Rainfall can seep into the
dam body, foundation rock, and bank slope, which will influence the dam seepage. The relationship
between rainfall and groundwater level is complex, and is related to rainfall and rainfall pattern,
infiltration conditions, topography, and geological conditions, and there is a certain hysteresis. The
average rainfall was used as a factor to simplify the simulated rainfall component, as shown below:

4
X
yP = bi (Pi − P0i ) (5)
i=1

where Pi (i = 1–4) is the rainfall of the monitoring day, 1 day before the monitoring day, the average
rainfall of 2 to 4 days before the monitoring day, and 5 to 8 days before the monitoring day; P0i (i = 1–4)
is the average rainfall corresponding to each of the above periods on the initial monitoring day; and bi
(i = 1–4) is the regression coefficient of the rainfall component.

2.2.3. Temperature Component yT

The temperature component is the variation of seepage caused by the temperature change of
the dam concrete and foundation rock. Thermal expansion will decrease the crack gaps, enhance
the impermeability, and then relieve seepage, while cooling shrinkage will increase the crack gaps,
undermine the impermeability, and then intensify seepage. The temperature of the dam body and
foundation rock varies periodically with atmospheric temperature, which can be expressed with
a periodic function. Considering the linear relationship between seepage and temperature, the
multi-period harmonic was chosen as the factor to represent the temperature component:

2 
2πit 2πit0 2πit 2πit0
X 
yT = c1i (sin − sin ) + c2i (cos − cos ) (6)
365 365 365 365
i=1

where c1i and c2i (i = 1–2) are the regression coefficients, respectively; t is the cumulative number of
days from the monitoring day to the initial monitoring day; and t0 is the cumulative number of days
from the first monitoring day of the data sequence taken by the modeling to the initial monitoring day.

2.2.4. Time Effect Component yθ

The time effect component is the variation of seepage caused by the change of time. The main
factors affecting the time effect component are the viscous flow of the dam concrete, and creep of the
rock foundation, which are often slow and undetectable, but the accumulation of such slow deformation
causes the stress of the foundation to adjust with time and has a great impact on the distribution of
dam foundation cracks. The influence of these factors on seepage flow is a slow time effect process,
and it is difficult to obtain accurate expressions by theoretical analysis at present. Typically, variation
Water 2019, 11, 1499 5 of 16

of the time effect component usually changes sharply at the beginning, and gradually turns stable
during the later period. A formula with a combination of two empirical formulas (linear function and
logarithmic function) represents the mathematical model of the general variation law of the time effect
component, which can be expressed as follows:

Water 2019, 11, x FOR PEER REVIEW yθ = d1 (θ − θ0 ) + d2 (lnθ − lnθ0 ) 5 of(7)

16

where d1 and d2 are regression coefficients; θ is the number of days from the initial monitoring day
𝑦 = 𝑑 (𝜃 − 𝜃 ) + 𝑑 (𝑙𝑛 𝜃 − 𝑙𝑛𝜃 ) (7)
divided by 100; and θ0 is the number of days from the first monitoring day of the data sequence
where 𝑑by and
divided 100, 𝑑 words, θ =coefficients;
are regression
in other 𝜃 is 0the
0.01t, θ0 = 0.01t . number of days from the initial monitoring
day divided by 100;
In summary, theand 𝜃 is monitoring
seepage the numbermodel
of dayscan
from the first monitoring
be expressed as follows:day of the data sequence
divided by 100, in other words, 𝜃 = 0.01t，𝜃 = 0.01𝑡" . #
5 4 2 7
In ŷsummary,
= a +the seepage
a (h − h monitoring
a h model
i − h can
i or be expressed
a h i − has ifollows:
P P   P   P  
t 0 i i )or 0i 4 i u1 u0 2 +
i u1 a hu0 7 −h i u(i−2) u0
i=1 i=1 𝑖 𝑖
i=1 𝑖 𝑖
i=3
𝑦 =𝑎 + P 𝑎 (ℎ − ℎ ) 𝑜𝑟
4 𝑎𝑖 ℎ𝑢1 − ℎ𝑢0 𝑜𝑟 𝑎𝑖 ℎ𝑢1 − ℎ𝑢0 + 𝑎𝑖 ℎ𝑢(𝑖−2) − ℎ𝑢0
+ bi (Pi − P0i𝑖=1) 𝑖=1 𝑖=3
4 i=1 (8)
2 h
+
+ 𝑏𝑖 (𝑃c𝑖1i−(sin
𝑃0𝑖 )2πit 2πit 2πit
i
− sin 3650 ) + c2i (cos 2πit
P
365 365 − cos 3650 )
𝑖=1
i=1 (8)
− θ2𝜋𝑖𝑡
+ +d1𝑐(θ(𝑠𝑖𝑛 0) +−d
2𝜋𝑖𝑡 lnθ ) 2𝜋𝑖𝑡
2 (lnθ −
𝑠𝑖𝑛 ) + 𝑐 0(𝑐𝑜𝑠 − 𝑐𝑜𝑠
2𝜋𝑖𝑡
)
365 365 365 365
where ŷt is the fitting value of the seepage monitoring data; and a0 is a constant.
+𝑑 (𝜃 − 𝜃 ) + 𝑑 (𝑙𝑛 𝜃 − 𝑙𝑛𝜃 )
Seepage monitoring values yt are variable, and the fitting values ŷt obtained by regression
where 𝑦will
analysis is deviate
the fitting value
from the of the seepage
monitoring monitoring
value to a certain data; 𝑎 is aestablishment
and After
extent. constant. of the seepage
Seepage
monitoring monitoring
model, values
the accuracy 𝑦 regression
of the are variable, andcan
fitting thebefitting values
measured 𝑦 obtained
by residual by regression
standard deviation
analysis
(S) will
and the deviatecorrelation
complex from the monitoring
coefficientvalue
(R). to a certain extent. After establishment of the seepage
monitoring model, the accuracy of the regression s fitting can be measured by residual standard
deviation (S) and the complex correlation coefficient
P n (R). 2
t=1 ( yt − ŷt )
S= (9)
∑ n −𝑦 k−−𝑦1
𝑆= 𝑡 (9)
tP 𝑛 − 𝑘 − 1 2
v
n
t=1 ( ŷt − yt )
R=
𝑅=
P∑n 𝑦𝑡 − 𝑦 2 (10)
(10)
(y − y )
∑t=1(𝑦 t− 𝑦 )t

where nn is
where is the
the number
number of
of monitoring data, kk is
monitoring data, is the
the number
number of
of independent variables, and
independent variables, yt is
and 𝑦 is the
the
average value of monitoring data.
average value of monitoring data.
2.3. Safety Class Division Standard
2.3. Safety Class Division Standard
The working state of seepage can be graded with the Pauta criterion [27]. It is considered that
The working state of seepage can be graded with the Pauta criterion [27]. It is considered that
the fitting curve calculated by the model is a reasonable value, and the confidence interval class is
the fitting curve calculated by the model is a reasonable value, and the confidence interval class is
determined according to a certain probability. The data in this interval are considered to be a reasonable
determined according to a certain probability. The data in this interval are considered to be a
fluctuation of seepage. Any error exceeding the range is not a random error but a gross error, and the
reasonable fluctuation of seepage. Any error exceeding the range is not a random error but a gross
data containing this error are determined to be in an abnormal state. The interval distributions of three
error, and the data containing this error are determined to be in an abnormal state. The interval
safety classes of the dependent variables are shown in Figure 2.
distributions of three safety classes of the dependent variables are shown in Figure 2.

Figure 2. Interval distribution diagram of three safety classes of dependent variables.

Figure 2. Interval distribution diagram of three safety classes of dependent variables.

When |𝑦 − 𝑦 | 2𝑆, it belongs to class I, indicating that the working state is normal;
When 2𝑆 |𝑦 − 𝑦 | 3𝑆, it belongs to class II, indicating that the working state is basically
normal. If there is an obvious tendency variation, it indicates that the working state is abnormal;
When |𝑦 − 𝑦 | 3𝑆, it belongs to class III, indicating that the working state is dangerous.
Water 2019, 11, 1499 6 of 16

 
When  yt − ŷt  ≤ 2S,
 it belongs to class I, indicating that the working state is normal;
When 2S <  yt − ŷt  ≤ 3S, it belongs to class II, indicating that the working state is basically normal.

If there is an obvious  tendency variation, it indicates that the working state is abnormal;
When  yt − ŷt  > 3S, it belongs to class III, indicating that the working state is dangerous.

The tendency variation is mainly reflected in the change of the time effect component. Positive
and negative values of dyθ /dt indicate the increase and decrease of the time effect component, and the
size of d2 yθ /dt2 indicates the rate of increase and decrease of the time effect component; dyθ /dt and
d2 yθ /dt2 can be used to achieve judgment of the variation tendency [28].
Water 2019, 11, x FOR PEER REVIEW 6 of 16

3. Multi-Indicator Grey Clustering Comprehensive Analysis

3. Multi-Indicator Grey Clustering Comprehensive Analysis
To deal with uncertainty problems during comprehensive evaluation with multi-indicators, grey
To deal with uncertainty problems during comprehensive evaluation with multi-indicators, grey
system theory is often adopted, which is used to solve the problem of uncertain systems, small samples,
system theory is often adopted, which is used to solve the problem of uncertain systems, small
and poor information [29]. The grey system studies the structure and function of a “black-grey-white”
samples, and poor information [29]. The grey system studies the structure and function of a “black-
box [30] through the organic relations and change rules among the object, element, and environment,
grey-white” box [30] through the organic relations and change rules among the object, element, and
as shown in Figure 3.
environment, as shown in Figure 3.

Figure3.3.Concept
Figure Conceptmap
mapof
ofthe
thegrey
greysystem.
system.
3.1. Grey Clustering Analysis Theory
3.1. Grey Clustering Analysis Theory
Grey clustering is a method of dividing monitoring indicators or monitoring objects into several
Grey clustering is a method of dividing monitoring indicators or monitoring objects into several
definable categories according to the grey correlation matrix or the grey whitenization weight function.
definable categories according to the grey correlation matrix or the grey whitenization weight
A cluster can be seen as a collection of monitoring objects belonging to the same class. This theory
function. A cluster can be seen as a collection of monitoring objects belonging to the same class. This
extracts valuable information by generating and mining some known information to calculate the grey
theory extracts valuable information by generating and mining some known information to calculate
correlation between different indicators and simplify complex systems [31].
the grey correlation between different indicators and simplify complex systems [31].
3.1.1. Basic Model
3.1.1. Basic Model
Suppose that xij (i = 1, 2, · · · , n; j = 1,2, · · · , m) is the monitoring value of object i about indicator j,
and f jkSuppose 2, · · ·𝑥, m;(ik==1,1,2,2,⋯,
(·) (j = 1,that · · ·n; the⋯,
, s)j =is1,2, m) is the
definite monitoring
weighted value
function of object
of the class ki about
about indicator
indicator j.j,
andgrey
The 𝑓 (·)clustering
(j = 1, 2, ⋯, m; k = 1, 2,
coefficient of⋯,
thes)object
is theidefinite
belonging weighted function
to the grey classof the class k about indicator j.
k is:
The grey clustering coefficient of the object i belonging to the grey class k is:
Xm  
σki = f jk xij ·η j (11)
𝜎 = 𝑓 𝑥 𝜂 (11)
j=1

whereη j𝜂is is
where thethe
weight of of
weight grey class
grey k about
class indicator
k about j. j.
indicator

3.1.2.
3.1.2.Modeling
ModelingSteps
Steps
(1)
(1)According
Accordingtotothetheevaluation
evaluation requirements,
requirements, thethe
indicator j is jdivided
indicator s grey
intointo
is divided classes,
s grey and the
classes, and
range
the rangeof values of each
of values indicator
of each is also
indicator is divided into sinto
also divided grey classes.
s grey For example,
classes. the value
For example, range
the value [a1 ,
range
a[a1, ]aof
s+1 s+1]the indicator
of the j is divided
indicator j is divided s small
intointo intervals:
s small intervals:
[a1, a2], [a2, a3], ⋯, [as−1, as], [as, as+1]
[a1 , a2 ], [a2 , a3 ], · · · , [as−1 , as ], [as , as+1 ]
(2) Calculation of the geometric midpoint between each small interval, 𝜆k = (ak + ak+1)/2, k = 1, 2,
⋯, s;
(3) Assume that the whitenization weight function value of 𝜆k belonging to the grey class k is 1.
Connect the point (𝜆k, 1) and the geometric midpoint (𝜆k−1, 0) of the grey class k − 1 to obtain the
triangular definite weighted functions 𝑓 (·), j = 1, 2, ⋯, m; k = 1, 2, ⋯, s. For 𝑓 (·) and 𝑓 (·), the
indicator j can be extended to the left and right to a0, as+2 respectively. For a monitoring value x of the
indicator j, it can be obtained as follows:
Water 2019, 11, 1499 7 of 16

(2) Calculation of the geometric midpoint between each small interval, λk = (ak + ak+1 )/2, k = 1, 2,
· · · , s;
(3) Assume that the whitenization weight function value of λk belonging to the grey class k is
1. Connect the point (λk , 1) and the geometric midpoint (λk−1 , 0) of the grey class k − 1 to obtain
the triangular definite weighted functions f jk (·), j = 1, 2, · · · , m; k = 1, 2, · · · , s. For f j1 (·) and f js (·), the
indicator j can be extended to the left and right to a0 , as+2 respectively. For a monitoring value x of the
indicator j, it can be obtained as follows:


 0, x < [ak−1 , ak+2 ]
x−ak−1

k ∈ [ak−1 , λk ]

f j (x) =  λk −ak−1 , x (12)

ak+2 −x

∈ [λk , ak+2 ]

ak+2 −λk , x


Water 2019, 11, x FOR PEER REVIEW 7 of 16

The membership degree f jk (x) of the grey class k (k = 1, 2, · · · , s) is calculated, as shown in Figure 4
The membership degree 𝑓 (x) of the grey class k (k = 1, 2, ⋯, s) is calculated, as shown in Figure
(The letters in the figure are consistent with the previous variables).
4 (The letters in the figure are consistent with the previous variables).

Figure 4. Schematic diagram of the endpoint triangle whitenization weight function.

Figure 4. Schematic diagram of the endpoint triangle whitenization weight function.

(4) In the process of multi-indicator grey clustering comprehensive analysis, machine learning is
(4) In the process of multi-indicator grey clustering comprehensive analysis, machine learning
used without human intervention. Therefore, the objective weighting method is used to determine the
is used without human intervention. Therefore, the objective weighting method is used to determine
weights of indicators. Entropy weight method is a method of calculating indicator weight based on
the weights of indicators. Entropy weight method is a method of calculating indicator weight based
comprehensive consideration of the information provided by various factors [32]. It calculates the
on comprehensive consideration of the information provided by various factors [32]. It calculates the
information entropy of each indicator to determine its difference, and then determines the importance
information entropy of each indicator to determine its difference, and then determines the
of the indicator in the comprehensive evaluation.
importance of the indicator in the comprehensive evaluation.
If there are n objects and m indicators, the prototype monitoring indicator values are normalized
If there are n objects and m indicators, the prototype monitoring indicator values are normalized
to obtain the matrix Sij = [sij ], and sij is the normalized value of the indicator j of the object i. According
to obtain the matrix 𝑆 = [𝑠 ], and 𝑠 is the normalized value of the indicator j of the object i.
to the information entropy definition, the information entropy of the indicator j is:
According to the information entropy definition, the information entropy of the indicator j is:
n
1
((q 𝑞== 1 j 𝑗==1,1,·⋯
X
P j𝑃==−q
−𝑞 s𝑠ij lnslnij𝑠 m)
· · , ,𝑚) (13)
(13)
lnn 𝑙𝑛 𝑛
i=1

The entropy
The entropy weight
weight of
of indicator
indicator jj is:
is:

1−𝑃 m
𝑒 = 1 − Pi (0 𝑒 1, X𝑒 = 1) (14)
ej = 𝑚 Pm∑ 𝑃
− (0 ≤ e j ≤ 1, e j = 1) (14)
m − j=1 Pi j=1
An indicator weight matrix A = (e1, e2, ⋯, ej) of the dam safety comprehensive early warning
systemAnbased
indicator on the entropy
weight matrix A =method
weight (e1 , e2 , · is
· · ,constructed.
ej ) of the dam safety comprehensive early warning
system(5)based
The comprehensive
on the entropyclustering
weight method coefficients of the object i (i = 1, 2, ⋯, n）about the grey class
is constructed.
k (k =(5) 2, ⋯,
1, The s) are calculated,
comprehensive as shown
clustering in Equation
coefficients object i (i = 1, 2, · · · , n) about the grey class k
(11).
of the
∗
From max 𝜎 o =
(k = 1, 2, · · · , s) are ncalculated,
(6) 𝜎 as shown in Equation (11). to grey class 𝑘 ∗ ; when there are multiple
, judge that object i belongs
(6) belonging
objects From max to σi the
k = σgrey
k∗
i
, judge 𝑘 ∗ , object
classthat i belongs toorgrey
the superiority class k∗ ; of
inferiority when
the there arebelonging
objects multiple objects
to the
1≤k≤s
∗
grey classto𝑘the
belonging can beclass
grey k∗ , the
further determined
superiorityaccording to the
or inferiority size
of the of the
objects comprehensive
belonging class k∗
clustering
to the grey
coefficient.
can be further determined according to the size of the comprehensive clustering coefficient.

3.2. Multi-Indicator Grey Clustering Comprehensive Analysis of Concrete Dam Seepage

The multi-indicator grey clustering comprehensive analysis model is here combined with the
seepage monitoring model and grey cluster analysis theory, which can realize a comprehensive
evaluation of seepage state. The first step of the model is to check the data of the monitoring indicators
and environmental variables, which are collected by the dam seepage monitoring system. The
Water 2019, 11, 1499 8 of 16

3.2. Multi-Indicator Grey Clustering Comprehensive Analysis of Concrete Dam Seepage

The multi-indicator grey clustering comprehensive analysis model is here combined with the
seepage monitoring model and grey cluster analysis theory, which can realize a comprehensive
evaluation of seepage state. The first step of the model is to check the data of the monitoring indicators
and environmental variables, which are collected by the dam seepage monitoring system. The seepage
monitoring model is used to analyze and fit the monitoring data of each indicator, and the trend of
each indicator is then predicted through mathematical expression of the model. Finally, grey clustering
analysis theory is used to centralize multiple indicators and realize the comprehensive evaluation of
seepage state.
Water 2019, ThePEER
11, x FOR overall framework of steps is shown in Figure 5.
REVIEW 8 of 16

Figure
Figure 5. Overallframework
5. Overall frameworkof of
thethe multi-indicator
multi-indicator grey
grey clustering
clustering comprehensive
comprehensive analysis
analysis of
of dam
dam seepage.
seepage.
4. Case Study
4. Case Study
The Cotton Beach Waterpower Project is located in Fujian Province, China. The project belongs
TheI. Cotton
to class Beachmulti-purpose
It is a large, Waterpower Project
projectiswith
located in Fujian
the main Province,
purpose of powerChina. The project
generation, and belongs
other
to class I. It is a large, multi-purpose project with the main purpose of
functions including flood control, navigation, and aquaculture. The maximum dam height is 111.0power generation, and other
m,
functions including flood control, navigation, and aquaculture. The maximum
the dam crest elevation is 179.0 m, and the total length of the dam crest is 308.5 m. The normal water dam height is 111.0
m, the
level ofdam crest elevation
the reservoir is 173 m,is 179.0 m, and
and the the capacity
storage total length of the
is 1.122 dam m
billion crest is 308.5
3 . The checkm.flood
The level
normalis
water m,
177.80 leveland ofthe
thetotal
reservoir
storageis 173 m, and
capacity the storage
is 2.035 billioncapacity
m3 . Theisdead
1.122 billion
water m is
level
3 . The
146check
m. Theflood
damlevel
has
is 177.80
seven dam m,sections:
and the total storage
the first and capacity
second damis 2.035 billion
sections aremleft
3. The dead water level is 146 m. The dam
bank retaining dam sections, the third
and fourth dam sections are overflow dam sections, and the fifth,left
has seven dam sections: the first and second dam sections are bank
sixth, andretaining
seventhdam damsections, the
sections are
third and fourth dam sections are overflow dam sections, and the fifth, sixth,
right bank retaining dam sections. The installed capacity is 600,000 kW. In order to ensure the seepage and seventh dam
sections
state aredam,
of the right abank retaining dam
comprehensive sections. The
monitoring installed
project capacity
is arranged is 600,000
inside of thekW. In buildings.
main order to ensure
The
the seepage state of the dam, a comprehensive monitoring project is arranged
information of monitoring points is shown in Table 1, and the location distribution of the monitoring inside of the main
buildings. The information of monitoring points is shown in Table 1, and the location distribution of
the monitoring points is shown in Figure 6. The monitoring points of seepage around the dam are
arranged in the seepage observation holes on both sides of the rock mass, and there are four
monitoring points on both the left and right sides of the dam (L1, R1 are in front of the dam curtain
and the others are behind the dam curtain).
Water 2019, 11, 1499 9 of 16

points is shown in Figure 6. The monitoring points of seepage around the dam are arranged in the
seepage observation holes on both sides of the rock mass, and there are four monitoring points on both
the left and right sides of the dam (L1, R1 are in front of the dam curtain and the others are behind the
dam curtain).

Table 1. The information of monitoring points.

Monitoring Object Number of Monitoring Points Monitoring Instrument

UP1, UP3, UP5, UP7, UP9, UP11, UP13, Pressure measuring
Dam seepage pressure
UP15, UP16 tube/osmometer
Dam leakage WE1, WE2, WE3, WE4, WE7, WE10 Measuring weir/osmometer
Water 2019, 11, x FOR PEER REVIEW 9 of 16
Seepage around dam L1, L3, L5, L7, R1, R3, R5, R6 Measuring weir/osmometer

1#dam section 2#dam section 3#dam section 4#dam section 5#dam section 6#dam section

179.00

UP16
UP1
140.00
WE10

UP15 UP3

UP5
100.00
WE7
UP13

Graphic symbols Code name Instrument name

UP11 76.00
UP7 UP Pressure measuring tube / osmometer
WE3 WE4 WE1WE2
68.00 Measuring weir/ osmometer
WE
UP9

Figure 6. Location distribution map of the dam seepage monitoring points.

Figure 6. Location distribution map of the dam seepage monitoring points.
4.1. Calculation of Seepage Monitoring Model
4.1. Calculation of Seepage Monitoring Model
Step 1: Collection of data from monitoring indicators and environmental factors from January 1,
2004Step to October 1: Collection 31, 2008 of through
data from themonitoring
dam seepage indicatorsmonitoring and environmental
system. factors from January 1,
2004 to October 31, 2008 through the dam seepage
Step 2: Substitution of these data into Equation (8) of the seepage monitoring model. By monitoring system.
Step
stepwise regression 2: Substitution analysis,of these the dataregressioninto Equationcoefficients (8) of were thesolved,
seepageand monitoring
the expression model.ofBy thestepwise
seepage
regression analysis,
monitoring model was obtained. the regression coefficients were solved, and the expression of the seepage
monitoring Step 3:model The data wasof obtained.
environmental factors from November 1, 2008 to December 31, 2008 were
Step 3: The
substituted into the calculated data of environmental
seepage monitoring factors from model, November
and the model 1, 2008output to Decembervalue of the 31, monitoring
2008 were
substituted into the calculated seepage
points was predicted. Figure 7 shows the variation curves between the monitoring values andmonitoring model, and the model output value of the the
monitoring points was
model output values of the monitoring points. predicted. Figure 7 shows the variation curves between the monitoring values
and the From model the output
variation valuescurves of the and monitoring
the multiple points. correlation coefficients of the monitoring values
water level of hole/m
and the model UP1
output values, it can
143.00be seen that UP3 seepage monitoring
the 114.10 model fitUP5 the monitoring
163.00
142.00 113.80
sequence well, the trends were consistent,
160.50
141.00 and the values were similar. The model values can be used
113.50
158.00
140.00
as155.50reasonable values of the monitoring 139.00
sequence to discriminate the existence
113.20 of the abnormal values
112.90
of the large offset data in the monitoring
153.00
138.00
137.00
sequence, so as to realize class 112.60division of the working states
150.50
of148.00the seepage indicators, R=0.956according
S=0.540 to
136.00
135.00
the differences of the
R=0.974 confidence
S=0.313 probability. And the
112.30

112.00
weight
R=0.961 S=0.073 of
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1
each monitoring point is calculated with the entropy weight method. The classification and 2008/1/1
weight 2009/1/1
of
each indicator areUP7 shown in Tables 100.00 2–4. UP9 UP11
110.00 120.00
108.00 115.00
95.00
106.00 110.00
104.00 90.00
105.00
102.00 85.00 100.00
100.00 95.00
80.00
98.00 90.00
96.00 75.00
R=0.776 S=1.379 R=0.979 S=1.096 85.00
94.00
R=0.974 S=1.513
70.00 80.00
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

UP13 UP15 UP16

112.50 165.00 155.50
160.00
110.00 155.00
155.00
107.50 150.00 154.50

105.00 145.00 154.00

140.00
 Water 2019, 11, 1499 10 of 16
10 of 16

water level of hole/m

163.00
UP1 143.00
UP3 114.10
UP5
142.00 113.80
160.50
141.00
113.50
158.00
140.00
113.20
155.50 139.00
112.90
138.00
153.00
137.00 112.60
150.50 112.30
136.00
R=0.956 S=0.540 R=0.974 S=0.313 R=0.961 S=0.073
148.00 135.00 112.00
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

UP7 UP9 UP11

110.00 100.00 120.00
108.00 115.00
95.00
106.00 110.00
90.00
104.00 105.00
102.00 85.00 100.00
100.00 95.00
80.00
98.00 90.00
96.00 75.00
R=0.776 S=1.379 R=0.979 S=1.096 85.00
94.00
R=0.974 S=1.513
70.00 80.00
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

UP13 UP15 UP16

112.50 165.00 155.50
160.00
110.00 155.00
155.00
107.50 150.00 154.50

105.00 145.00 154.00

140.00
102.50 153.50
135.00
R=0.919 S=0.487 R=0.868 S=1.772 R=0.854 S=0.113
100.00 130.00 153.00
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

seepage flow/L/s
2.00 WE1 0.80
WE2 2.50 WE3
1.50 2.00
0.60
1.00 1.50

0.50 0.40 1.00

0.00 0.50 R=0.983 S=0.076

0.20
-0.50 0.00
R=0.869 S=0.246 R=0.807 S=0.063
-1.00 0.00 -0.50
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

6.00 WE4 1.00

WE7 1.50
WE10

4.00 1.00
0.50

2.00 0.50

0.00
0.00 0.00

R=0.939 S=0.376 R=0.807 S=0.111 R=0.868 S=0.200

-2.00 -0.50 -0.50
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

water level of hole/m

200.00
L1 170.00
L3 146.00
L5

160.00
190.00 145.00

150.00

180.00 144.00
140.00

R=0.867 S=2.487 R=0.927 S=2.323 R=0.701 S=0.088

170.00 130.00 143.00
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

92.00 L7 185.00 R1 180.00

R3
90.00
175.00 170.00
88.00
165.00 160.00
86.00

155.00 150.00
84.00

R=0.820 S=0.454 R=0.928 S=2.924 R=0.930 S=2.695

82.00 145.00 140.00
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

180.00
R5 170.00
R6
170.00 160.00

160.00 150.00

150.00 140.00

140.00 130.00

130.00 120.00
R=0.720 S=2.325 R=0.796 S=2.901
120.00 110.00
2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1 2004/1/1 2005/1/1 2006/1/1 2007/1/1 2008/1/1 2009/1/1

278 Figure 7.
Figure 7. The variation curves between the monitoring values and the mode output values.
values

279 From the variation curves and the multiple correlation coefficients of the monitoring values and
280 the model output values, it can be seen that the seepage monitoring model fits the monitoring
281 sequence well, the trends are consistent and the values are similar. It can be used as reasonable values
282 of the monitoring sequence to discriminate the existence of the abnormal values of the large offset
Water 2019, 11, 1499 11 of 16

Table 2. The classification results and weights of dam seepage pressure.

Class UP1 UP3 UP5 UP7 UP9 UP11 UP13 UP15 UP16
Class I [0, 1.08] [0, 0.63] [0, 0.15] [0, 2.76] [0, 2.19] [0, 3.03] [0, 0.97] [0, 3.54] [0, 0.23]
Class II (1.08, 1.62] (0.63, 0.94] (0.15, 0.22] (2.76, 4.14] (2.19, 3.29] (3.03, 4.54] (0.97, 1.46] (3.54, 5.32] (0.23, 0.34]
Class III (1.62, +∞) (0.94, +∞) (0.22, +∞) (4.14, +∞) (3.29, +∞) (4.54, +∞) (1.46, +∞) (5.32, +∞) (0.34, +∞)
Information entropy 61.902 60.892 66.481 60.338 57.992 54.404 36.561 9.097 9.094
Weight 0.071 0.069 0.076 0.069 0.066 0.062 0.041 0.009 0.009

Table 3. The classification results and weights of dam leakage.

Class WE1 WE2 WE3 WE4 WE7 WE10

Class I [0, 0.49] [0, 0.13] [0, 0.15] [0, 0.75] [0, 0.22] [0, 0.40]
Class II (0.49, 0.74] (0.13, 0.19] (0.15, 0.23] (0.75, 1.13] (0.22, 0.33] (0.40, 0.60]
Class III (0.74, +∞) (0.19, +∞) (0.23, +∞) (1.13, +∞) (0.33, +∞) (0.60, +∞)
Information entropy 8.464 12.763 4.495 9.217 12.511 10.045
Weight 0.009 0.014 0.005 0.010 0.013 0.010

Table 4. The classification results and weights of seepage around dam.

Class L1 L3 L5 L7 R1 R3 R5 R6
Class I [0, 4.97] [0, 4.65] [0, 0.18] [0, 0.91] [0, 5.85] [0, 5.39] [0, 4.65] [0, 5.80]
Class II (4.97, 7.46] (4.65, 6.97] (0.18, 0.26] (0.91, 1.36] (5.85, 8.77] (5.39, 8.09] (4.65, 6.98] (5.80, 8.70]
Class III (7.46, +∞) (6.97, +∞) (0.26, +∞) (1.36, +∞) (8.77, +∞) (8.09, +∞) (6.98, +∞) (8.70, +∞)
Information entropy 42.261 36.613 59.311 66.776 46.283 45.546 74.592 39.119
Weight 0.048 0.041 0.068 0.076 0.053 0.052 0.085 0.044
Water 2019, 11, 1499 12 of 16
Water 2019, 11, x FOR PEER REVIEW 12 of 16

4.2. Comprehensive Clustering Calculation of Multiple Dependent Variables

4.2. Comprehensive Clustering Calculation of Multiple Dependent Variables
The whitenization weight function is the basis of grey clustering. The membership function of
fuzzyThe whitenization
evaluation weight
can reflect function is the
the relationships basis of
between grey clustering.
clustering The
indicators andmembership
grey classes.function of
Since the
fuzzy evaluation can reflect the relationships between clustering indicators and grey classes.
evaluation factors of monitoring sequence indicators have a lower limit and no upper limit, the upper Since
the lower
and evaluation
limit factors of monitoring
whitenization sequencewas
weight function indicators
used, ashave
showna lower limit8.and no upper limit, the
in Figure
upper and lower limit whitenization weight function was used, as shown in Figure 8.

Figure8.8.Schematic
Figure Schematicdiagram
diagramof
ofthe
thedefinite
definiteweighted
weightedfunction.
function.

The
Themulti-indicator
multi-indicatorgrey
greyclustering
clusteringcomprehensive
comprehensiveanalysis
analysismodel
modelwas
wasused
usedon
onthe
theperiod
periodfrom
from
November
November 1,1, 2008
2008 to
to December
December  31,
31, 2008
2008
 to
to forecast
forecast the
thechange
changetrend
trendofofseepage,
seepage,combining
combining the
the
the y|𝑦
monitoring  t − ŷt /S of each monitoring points and the clustering coefficients of
monitoringdata
datatotocalculate
calculatethe − 𝑦 |/𝑆 of each monitoring points and the clustering coefficients
each seepage
of each indicator,
seepage as shown
indicator, as shownin Figure
in Figure9. 9.

4.3. Results and

|𝑦𝑡−𝑦 Discussion
t̂ |/S of dam seepage pressure monitoring point
Clustering coefficient change histogram of dam
seepage pressure
3 1.00
Each indicator has three grey class clustering coefficients; the bigger the clustering coefficient of
UP1 0.95
the grey class, the more likely the working stateUP3of the
2.5
0.90
indicator will be in that grey class. The larger
the clustering
2
coefficient of grey class I is, the better
UP5 0.85the seepage state will be. On the contrary, the

larger the clustering coefficient of grey class III UP7 is, the
0.80worse the seepage state will be. The sum of Ⅲ the
1.5 0.75
three grey classes at any time is 1. UP9
Ⅱ
UP11 0.70
From
1 Figure 9, it can be seen that the state of dam seepage pressure was basically in the stage Ⅰ of
UP13 0.65
normal operation of grey class I across these two months,
UP15 0.60
which were both greater than 0.75. After
0.5
December, the proportion of grey class II increased, UP16 0.55 reaching a maximum of 0.25, and dangerous

0
conditions of grey class III began to emerge, indicating0.50 that the state of dam seepage pressure tends to
2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31 2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31
develop towards the abnormal, which was mainly attributed to the large offset anomaly of monitoring
points UP9 |𝑦 and UP11.
𝑡−𝑦 𝑡̂ |/S of dam leakage monitoring point Clustering coefficient change histogram of dam
The
5 dam leakage is monitored manually, and only1 once a week, so the monitoring sequence is less
leakage

than automatic monitoring. It can be seen that the grey class I of dam leakage was basically greater
0.95
4 0.9
than 0.75, which would indicate normal operation. WE1 However, grey class II had an upward trend, and
0.85 Ⅲ
gradually
3 changed to grey class III, with a maximum WE2 of
0.8 0.23, indicating that the state of dam leakage
had a trend of dangerous development, whichWE3 was0.75mainly attributed to the large offset anomalyⅡof
2 0.7
monitoring points WE4 and WE3. WE4
0.65
Ⅰ
1
The grey class I of seepage around the dam was generally higher than 0.9, the maximum grey
WE7 0.6
0.55
class II
0
reached 0.1, and only monitoring point L3 WE10had a small offset anomaly, indicating that seepage
0.5
around dam was
2008/11/1 in normal
2008/11/16 2008/12/1 condition
2008/12/16 during
2008/12/31 these two months.
2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31

Figure 10 is a comprehensive clustering coefficient histogram of seepage, which takes into account
Clustering coefficient change histogram of seepage
|𝑦 −𝑦 𝑡̂ |/S of seepage around dam monitoring point
the above 𝑡three indicators. It shows that the dam seepage 1
was generally aroundindamthe stage of grey class
3
I during this period, and there was an upward L1
trend
0.95 of grey class II after December, reaching a
2.5
maximum of 0.13, indicating that the seepage was in0.9a good state of operation, but there was a small
L3
0.85
offset 2anomaly. L5
0.8
L7 Ⅲ
1.5 0.75
R1 Ⅱ
0.7
1 R3 Ⅰ
0.65
R5 0.6
0.5
R6 0.55
0 0.5
2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31 2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31

Figure 9. |𝑦 − 𝑦 |/𝑆 and the clustering coefficient of each seepage indicator.

 Figure 8. Schematic diagram of the definite weighted function.

The multi-indicator grey clustering comprehensive analysis model was used on the period from
November 1, 2008 to December 31, 2008 to forecast the change trend of seepage, combining the
monitoring 1499 to calculate the |𝑦 − 𝑦 |/𝑆 of each monitoring points and the clustering coefficients
Water 2019, 11,data 13 of 16
of each seepage indicator, as shown in Figure 9.
|𝑦𝑡−𝑦 t̂ |/S of dam seepage pressure monitoring point Clustering coefficient change histogram of dam
seepage pressure
3 1.00

UP1 0.95
2.5
UP3 0.90

UP5 0.85
2
UP7 0.80 Ⅲ
1.5 UP9 0.75
Ⅱ
UP11 0.70
Water12019, 11, x FOR PEER REVIEW 13 Ⅰ
of 16
UP13 0.65
UP15 0.60
0.5
4.3. Results and Discussion UP16 0.55
0 0.50
Each indicator
2008/11/1 2008/11/16 has three2008/12/16
2008/12/1 grey class clustering coefficients;
2008/12/31 2008/11/1 the bigger
2008/11/16 the clustering
2008/12/1 2008/12/16 coefficient
2008/12/31 of
the grey class, the more likely the working state of the indicator will be in that grey class. The larger
the clustering |𝑦𝑡−𝑦 𝑡̂ coefficient
|/S of dam leakage monitoring
of grey classpoint
I is, the better theClustering
seepage state will
coefficient changebe. On the
histogram contrary, the
of dam
5 leakage
larger the clustering coefficient of grey class III is, the 1 worse the seepage state will be. The sum of the
0.95
three4 grey classes at any time is 1. 0.9
WE1
From Figure 9, it can be seen that the state of dam 0.85 seepage pressure was basically in the stage Ⅲ of
3 WE2
normal operation of grey class I across these two months, 0.8 which were both greater than 0.75. After
WE3 0.75 Ⅱ
December,
2 the proportion of grey class II increased, 0.7 reaching a maximum of 0.25, and dangerous
WE4 Ⅰ
conditions of grey class III began to emerge, indicating 0.65 that the state of dam seepage pressure tends
1 WE7 0.6
to develop towards the abnormal, which was mainly attributed to the large offset anomaly of
0.55
WE10
monitoring
0 points UP9 and UP11. 0.5
2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31
The dam leakage is monitored manually, and 2008/11/1 only once2008/11/16
a week, so 2008/12/1 2008/12/16
the monitoring 2008/12/31
sequence is
less than automatic monitoring. It can
|𝑦𝑡−𝑦 𝑡̂ |/S of seepage around dam monitoring point
be seen that the grey class
Clustering I of
coefficient dam
change leakage
histogram of was
seepagebasically
greater
3 than 0.75, which would indicate normal operation. 1 However,around greydam class II had an upward
trend, and gradually changed to grey class III,L1with a maximum of 0.23, indicating that the state of
0.95
2.5 0.9
dam leakage had a trend of dangerous development, L3 which was mainly attributed to the large offset
0.85
2
anomaly of monitoring points WE4 and WE3. L5
0.8
L7 Ⅲ
1.5The grey class I of seepage around the dam was 0.75 generally higher than 0.9, the maximum grey
R1 Ⅱ
class II reached 0.1, and only monitoring point L3 had 0.7 a small offset anomaly, indicating that seepage
Ⅰ
1 R3
0.65
around dam was in normal condition during these R5
two months.
0.6
0.5
Figure 10 is a comprehensive clusteringR6 coefficient 0.55
histogram of seepage, which takes into
account
0 the above three indicators. It shows that the
0.5 dam seepage was generally in the stage of grey
2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31 2008/11/1 2008/11/16 2008/12/1 2008/12/16 2008/12/31
class I during this period, and there was an upward trend of grey class II after December, reaching a
maximum of 0.13, 9.9.|𝑦 yt−−𝑦that
indicating ŷt|/𝑆
/Sthe
andseepage was in a good state of seepage
operation, but there was a small
 
Figure
Figure and the
the clustering
clustering coefficient
coefficient of
of each
each seepage indicator.
indicator.
offset anomaly.

Figure 10. Comprehensive

Comprehensive clustering
clustering coefficient
coefficient histogram of seepage.

In summary, although dam seepage was overall in normal operation, some abnormal conditions
were discovered through the multi-indicator grey clustering comprehensive analysis model. The data
of some monitoring points had obvious deviations from the model output values, such as UP9, UP11,
WE3, and WE4, which were distributed in sections 3# and 4# of the overflow dam section, indicating
Water 2019, 11, 1499 14 of 16

In summary, although dam seepage was overall in normal operation, some abnormal conditions
were discovered through the multi-indicator grey clustering comprehensive analysis model. The data
of some monitoring points had obvious deviations from the model output values, such as UP9, UP11,
WE3, and WE4, which were distributed in sections 3# and 4# of the overflow dam section, indicating
that the overflow dam section is the main section where seepage anomaly occurs. Because the overflow
dam section sometimes needs continuous drainage, it is easy for water flow to scour the dam and
negative pressure to appear on overflow surface, resulting in leakage and intestinal retention of fluid.
In order to ensure the normal operation of dam seepage, it is necessary to monitor abnormal parts
moving forward to prevent the occurrence of dangerous accidents.

5. Conclusions
The purpose of this paper was to construct a multi-indicator grey clustering comprehensive
analysis model for seepage safety monitoring of concrete dams, using the seepage monitoring model
to predict the change tendency of monitoring points, combined with grey cluster analysis theory to
fuse the multiple indicators, so as to realize a comprehensive evaluation of the dam seepage state. The
following conclusions can be obtained:
The seepage indicator system is complete and representative, and takes into account the main
factors affecting the seepage state of the dam, and each factor includes multiple monitoring points of
the whole dam section. The seepage monitoring model adopts different mathematical expressions
for different indicators according to the correlation between water level component and indicators.
Stepwise regression method is used to retain the environmental factors with high correlation and
eliminate the environmental factors with poor correlation, which makes the model fit the monitoring
data better. By using the Pauta criterion to classify the indicators according to a certain probability, the
problem of random errors in actual monitoring and statistics was eliminated.
Grey clustering analysis theory combined with the seepage monitoring model was applied to the
integration of multi-source information monitoring data of dam seepage, which is an interdisciplinary
practice. The evaluation results reflect the trend of dam seepage through the continuous change of
grey class with time series, which is a major improvement from static state to dynamic state evaluation.
There are many uncertainties in the fusion of multi-index and multi-information; grey clustering
analysis theory can find the grey correlations among the indicators, mine useful information from
limited data, and divide observation objects into several definable categories with a whitenization
weight function. This combination of different factors can simplify the complex system and is very
suitable for comprehensive evaluation of dam seepage. Through the calculation of an engineering case,
the validity and availability of the method was verified. In practical engineering applications, when
the calculation results are generally shown as grey class I, it indicated that the dam seepage was in the
normal operation stage. However, attention should be paid to the changing trends of the other grey
classes. If there is a tendency variation of offset anomaly, it indicates that the dam seepage state is
deteriorating gradually. When the calculation results reach grey class II, preventive measures must be
taken for abnormal parts, such as drainage decompression, curtain grouting, concrete filling, etc. The
seepage state of the dam is expressed quantitatively, which is convenient for the daily maintenance
of the dam and the early warning of the dangerous situation. This method can also be applied to
other engineering fields. As long as data with multiple evaluation indicators are available, valuable
information can be extracted through the multi-indicator grey clustering comprehensive analysis
model, and grey classification can then be carried out to achieve comprehensive evaluation of the
target object, which makes the evaluation work more quantitative and intuitive.

Author Contributions: Conceptualization, J.L. and X.C. Methodology, J.L. and X.C. Data Curation, J.L.
Writing—Original Draft Preparation, J.L. Writing—Review and Editing, J.L., X.C., C.G. and Z.H.
Funding: This research was funded by the National Natural Science Foundation of China under Grant No. 51609217
and Grant No. 51679222.
Water 2019, 11, 1499 15 of 16

Acknowledgments: Appreciation is due to all members of Li Zongkun’s research group of Zhengzhou University
for their valuable opinions on this study.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Fang, C.; Duan, Y. Statistical analysis of dam-break incidents and its cautions. Yellow River 2010, 42, 96–100.
[CrossRef]
2. Wang, H. Dam safety and monitoring. Water Resour. Hydropower Eng. 2009, 40, 126–132. [CrossRef]
3. Chen, X.; Gu, C.; Chen, H. Early warning of dam seepage with cooperation between principal component
analysis and least squares wavelet support vector machine. Fresenius Environ. Bull. 2013, 22, 500–507.
4. Shen, C.; Lu, Z. Factor Selection of Statistical Model for Seepage Observation of Concrete Dams. Large Dam
Saf. 1992, 4, 1–10.
5. Li, X.; Zhen, D.; Ju, Y. Input factor optimization study of dam seepage statistical model based on copula
entropy theory. J. Hohai Univ. 2016, 44, 370–376.
6. Zhang, Q.; Lin, C.; Xiang, W. Statistical regression analysis of seepage and seepage-pressure of gravity dam
foundation of Shiban Hydropower Station. Rock Soil Mech. 2006, 27, 1831–1834. [CrossRef]
7. Wei, B.; Gu, M.; Li, H.; Xiong, W.; Xu, Z. Modeling method for predicting seepage of RCC dams considering
time-varying and lag effect. Struct. Control Health Monit. 2018, 25, 1–14. [CrossRef]
8. Qiu, J.; Zhang, D.; Zhu, K. Seepage Monitoring Models Study of Earth-Rock Dams Influenced by Rainstorms.
Math. Probl. Eng. 2016, 2016, 1656738. [CrossRef]
9. Wang, X.; Yu, H.; Lv, P. Seepage Safety Assessment of Concrete Gravity Dam Based on Matter-Element
Extension Model and FDA. Energies 2019, 12, 502. [CrossRef]
10. Mei, Y.; Zhong, Y. Fuzzy extension evaluation model of dam seepage behavior based on entropy weight.
Water Resour. Power 2011, 29, 58–61. [CrossRef]
11. He, J.; Gao, Q.; Shi, Y. A multi-hierarchical comprehensive evaluation method of dam safety based on cloud
model. Syst. Eng. Theory Pract. 2016, 36, 2977–2983. [CrossRef]
12. Li, Z.; Li, Q. Comprehensive Assessment of Dam Risk Consequences Caused by the Dam Failure Based on
the Set Pair Analysis. Yellow River 2016, 38, 111–114. [CrossRef]
13. Bui, K.T.T.; Bui, D.T.; Zou, J.; Van Doan, C.; Revhaug, I. A novel hybrid artificial intelligent approach based
on neural fuzzy inference model and particle swarm optimization for horizontal displacement modeling of
hydropower dam. Neural Comput. Appl. 2018, 29, 1495–1506. [CrossRef]
14. Peyras, L.; Carvajal, C.; Felix, H.; Bacconnet, C.; Royet, P.; Becue, J.P.; Boissier, D. Probability-based assessment
of dam safety using combined risk analysis and reliability methods—Application to hazards studies. Eur. J.
Environ. Civ. Eng. 2012, 16, 795–817. [CrossRef]
15. Fan, X.; Wu, J.; Shi, P.; Zhang, X.; Xie, Y. A novel automatic dam crack detection algorithm based on
local-global clustering. Multimed. Tools Appl. 2018, 77, 26581–26599. [CrossRef]
16. He, Y.; Zhao, M.; Wang, K.; Liu, P. Fuzzy Comprehensive Evaluation of Earth Rockfill Dam Seepage Security
Risk Based on Cloud Model. Water Resour. Power 2018, 36, 83–86.
17. Talona, A.; Curt, C. Selection of appropriate defuzzification methods: Application to the assessment of dam
performance. Expert Syst. Appl. 2017, 70, 160–174. [CrossRef]
18. Liu, S.; Yang, Y. Advances in grey system research (2004–2014). J. Nanjing Univ. Aeronaut. Astronaut. 2015, 47,
1–8. [CrossRef]
19. Meng, W.; Liu, S.; Zeng, B.; Xie, N. Standard triangular whitenization weight function and its application in
grey clustering evaluation. J. Grey Syst. 2012, 24, 39–48.
20. Sun, W.; Li, L.; Zeng, H. Dam risk consequences comprehensive evaluation model based on the method of
gray correlation. Yellow River 2012, 34, 102–104. [CrossRef]
21. Hu, Y.; Shao, C.; Gu, C.; Meng, Z. Concrete Dam Displacement Prediction Based on an ISODATA-GMM
Clustering and Random Coefficient Model. Water 2019, 11, 714. [CrossRef]
22. Liu, Y. Dam Safety Evaluation Model & Its Application Based on Entropy Weight and Idea Point Method.
Water Resour. Power 2016, 34, 73–76.
23. Li, L.; Yang, B.; Wei, B. Cloud model of dam behavior safety diagnosis based on improved analytic hierarchy
process. J. Water Resour. Water Eng. 2018, 29, 209–214. [CrossRef]
Water 2019, 11, 1499 16 of 16

24. Tao, C.; Chen, H.; Kong, S. Seepage Analysis of Clay Core Dam Based on Prototype Monitoring. J. Yangtze
River Sci. Res. Inst. 2017, 34, 70–73. [CrossRef]
25. Fu, X.; Zheng, D.; Su, G. Application of Improved Matter Element Extension Evaluation Method to Dam
Seepage Assessment. J. China Three Gorges Univ. 2015, 37, 26–29. [CrossRef]
26. Wu, Z. Safety Monitoring Theory & Its Application of Hydraulic Structures; Higher Education Press: Beijing,
China, 1990; pp. 130–150.
27. Zhang, X.; Xi, Q.; Liu, S. Application of GNSS monitoring data pretreatment to the prediction of deformation
GNSS. Eng. Surv. Mapp. 2016, 25, 42–45. [CrossRef]
28. Gu, C.S.; Wei, B.W.; Xu, Z.K.; Liu, D.W. Fluid-solid coupling model based on endochronic damage for roller
compacted concrete dam. J. Cent. South Univ. 2013, 20, 3247–3255. [CrossRef]
29. Zhao, B.; Ren, Y.; Gao, D.; Xu, L. Performance ratio prediction of photovoltaic pumping system basedon grey
clustering and second curvelet neural network. Energy 2019, 171, 360–371. [CrossRef]
30. Liu, S.; Xie, N. Grey System Theory and Application; Science Press: Beijing, China, 2013; pp. 5–18.
31. Delgado, A.; Romero, I. Environmental conflict analysis using an integrated grey clustering and
entropy-weight method: A case study of a mining project in Peru. Environ. Model. Softw. 2016, 77,
108–121. [CrossRef]
32. Shemshadi, A.; Shirazi, H.; Toreihi, M.; Tarokh, M. A fuzzy VIKOR method for supplier selection based on
entropy measure for objective weighting. Expert Syst. Appl. 2011, 38, 12160–12167. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).