You are on page 1of 18

Natural Resources Research (Ó 2022)

https://doi.org/10.1007/s11053-022-10042-x

Original Paper

An Improved Connection Cloud Model of an Updated


Database: A Multicriteria Uncertainty Model for Coal
Burst Liability Evaluation

Jian Zhou ,1,2 Chao Chen,1 Cong Wei,1 and Kun Du1,2

Received 20 September 2021; accepted 25 February 2022

In the context of complex geological environments, coal burst liability (CBL) evaluation
struggles with uncertainty and fuzziness of indices and with obtaining satisfactory accuracy.
Therefore, an improved connection cloud model (ICCM) is proposed for CBL evaluation
based on set pair analysis referring to four indices. The numerical characteristics of the
ICCM, namely expectation (Ex), entropy (En), hyper-entropy (He) and parameter (f), were
determined by set pair analysis and classification standards. Then, the index measurement-
based entropy weight method and the analysis hierarchy process (AHP) were used to
measure clearly the importance of each index. Finally, Euclidean distance was employed to
classify the CBL objectively, making full use of the valuable information mined from the
data. An updated CBL database (of 147 data records) was constructed to validate the
potential of the proposed algorithm for CBL classification. Additionally, a graphical user
interface was developed to make the CBL evaluation visible. The classification metrics of the
ICCM, i.e., accuracy and Kappa coefficient, were 0.88 and 0.772, respectively, which were
better than those of the other conventional uncertainty methods. The results reveal that the
proposed ICCM is a prospective tool for uncertainty information treatment and subjectivity
removal in the field of coal burst control.
KEY WORDS: Coal burst, Bursting liability, Cloud model, Set pair analysis, Euclidean distance,
Classification.

INTRODUCTION baei et al., 2016), occur frequently due to the ad-


vance of deep mining. Coal burst is a common and
Currently, the scientific community is paying severe geological hazard that leads to casualties and
more attention to deep mining, as surface solid equipment damage.
mineral resources are exhausted (Li et al., 2017). Coal bursts result from the violent release of
However, various geological hazards, e.g., rock burst energy accumulated in the coal mass. Although
(Ortlepp, 2005; Zhou et al., 2016, 2018), coal burst various studies have been carried out on under-
(Mark, 2014; Zhou et al., 2021b), water inrush ground coal mines (e.g., Chen et al., 2021b; Li et al.,
(Wang et al., 2012), pillar instability (Li et al., 2021a, 2021e, 2021f; Sahoo et al., 2021; Tang et al., 2021;
2021b; Du et al., 2022) and roof collapse (Aghaba- Zhang et al., 2021), this disaster still cannot be
controlled effectively because of the complexity of
1
School of Resources and Safety Engineering, Central South geological conditions and the uncertainty of loca-
University, Changsha 410083, China. tion. A large variety of coal burst events have been
2
To whom correspondence should be addressed; e-mails: recorded in various countries in the last few years,
csujzhou@hotmail.com, j.zhou@csu.edu.cn, dukuncsu@csu.edu.cn

Ó 2022 International Association for Mathematical Geosciences


Zhou, Chen, Wei, and Du

e.g., the Czech Republic (Ortlepp, 2005), China (Pan


et al., 2013), the USA (Mark, 2014) and Poland
(Makówka, 2016). For example, on March 26, 2012,
a severe rock burst event, induced by roof fracturing,
occurred in the Xingan Coal Mine, Hegang City,
resulting in casualties and damage to underground
infrastructure (Wang et al., 2016). On March 27,
2014, the coal burst that occurred in the Qianqiu
Coal Mine killed six miners (Lu et al., 2015).
Additionally, the Yuejin Coal Mine (Xu, 2011),
Wulong Coal Mine (Qi et al., 2019) and Qianqiu
Coal Mine (Lu et al., 2015) also observed challeng-
ing geological hazards as coal resources were ex-
tracted.
Therefore, it is necessary and crucial to classify
coal burst liability (CBL) and offer valuable infor-
mation for decision-makers as much as possible. In Figure 1. Summary of current CBL studies.
this case, various theoretical methodologies, includ-
ing, among others, the Chinese national standard
(GB/T 25,217.2-2010), principal component analysis et al., 2020; Zhou et al., 2021, 2022; Dai et al., 2022),
(PCA) (Cai et al., 2016), comprehensive evaluation extreme gradient boosting (Ding et al., 2020; Zhou
(Cai et al., 2018), analytic hierarchy process (Chen et al., 2021f), artificial neural networks (Yu et al.,
et al., 2019) and unascertained measurement theory 2021), virtual field optimization methods (Zhou,
(Zhou et al., 2020a, 2021a; Chen, et al., 2021a), and et al., 2021) and support vector machines (Zhou
practical techniques, including, among others, mon- et al., 2021d, 2022b; Li et al., 2021c, 2021d), has
itoring techniques (Wei et al., 2018), water inrush improved further the reliability of engineering de-
(Wei et al., 2018) and destress blasting (Zhou et al., sign. The main purpose of this work was to propose
2020b) have been proposed in previous studies to an improved connection cloud model (ICCM) based
tackle the issue. on the classic cloud theory (Zhang et al., 2015; Zhou
Figure 1 summarizes the special distribution of et al., 2016) and the Euclidean distance incorporat-
countries, mechanisms, control techniques and ing set pair analysis for CBL classification. First, the
evaluation methodologies (Li et al., 2017; Zhang set pair theory was used to determine the numerical
et al., 2017; Wei et al., 2018; Sabapathy et al., 2019; characteristics of the conventional cloud model.
Du et al., 2020, 2021; Xie et al., 2021; Zhou et al., Then, a combination weight was employed to ana-
2021b) associated with CBL evaluation. However, lyze dynamically the importance of every index, and
there are still several issues that need to be resolved the Euclidean distance was applied to replace the
to improve the reliability of CBL assessment. First, widely used tool, namely the maximum measure-
there is a need to determine the importance of ment principle, as the classification standard.
individual indicators in the index system and to re- Eventually, the constructed model was utilized to
move the uncertainty caused by subjective judgment evaluate the CBL of the dataset collected from the
in a great manner. Second, more high-efficiency literature. The technique routine is shown in Fig-
classification criteria should be explored to reduce ure 2.
the loss of valuable information while making a
decision. Finally, due to the complexity and uncer-
tainty of on-site environments, as well as the devi- METHODOLOGY
ation of the collected dataset, it is important to deal
with the dataset collected considering the random- One-Dimensional Cloud Model
ness and uncertainty simultaneously.
In the last few years, the wide use of a large The cloud model proposed by Li et al. (2004)
volume of advanced intelligence algorithms in the aims to describe quantitatively the qualitative con-
fields of engineering, including random forests (Han cept, which can handle uncertainty information
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

Figure 2. Routine of the proposed CBL technique.

efficiently based on statistics methodology and fuzzy Improved Integration Cloud Model
set theory (Wang et al., 2020b).
Determination of the connection for the
Step 1 Suppose that T and G are the universe and
boundary of adjacent intervals by the conventional
qualitative concept of objects to be evaluated, and
cloud model may be inaccurate, justifying largely the
x is the cloud droplet generated through a specific
use of the ‘‘3En’’ principle (Wang et al., 2020a).
principle. Then, the connection degree l(x) can be
Thus, the connection cloud model was combined
determined as:
with connection number theory to quantify reliably
the membership degree associated with individual
lð xÞ : T ! ½0; 1; 8x 2 T ! lð xÞ ð1Þ indices.
Suppose that universe T consists of n alterna-
tives, and the characteristics of individual alterna-
Step 2 Determine the distribution characteristics of
tives are described through m-dimensional attributes
cloud droplet x by numerical characteristics (Ex,
Ti ¼ fm1 ; m2 ; . . . ; mm gð1  i  nÞ. The qualitative
En, He). Ex, the expected value of cloud droplet x
concept can be divided into Kðk ¼ 1; 2; . . . ; KÞ
on universe T, determines the location of concept
grades. If Ti ¼ fm1 ; m2 ; . . . ; mm g satisfies the normal
G. Entropy (En) and hyper-entropy (He) are
distribution fm1 ; m2 ; . . . ; mm g  N fExðExk1 ; Exk2 ;
associated with the width and thickness of cloud  0 0  2
droplets, respectively. . . . ; Ex Þ; En Enk1 ; En0k2 ; . . . ; En0km g and En0
 0 km 0 
Step 3: Generate cloud droplet (xi, lxi) ( i ¼ Enk1 ; Enk2 ; . . . ; En0km  N fEnðEnk1 ; Enk2 ; . . . ;
1; 2; . . .; N whereN is determined
 by experts),
 thus: Enkm Þ; ½HeðHek1 ; Hek2 ; . . . ; Hekm Þ2 g, then the mem-
02 02 2
xi  N Exi ; Eni and Eni  N Eni ; Hei .
Zhou, Chen, Wei, and Du

 
bership degree li mkj of index mj ð1  j  mÞ
belonging to grade k can be calculated as Hekj ¼ b ð10Þ
0   k1
  Xm mk  Exk fj
9  j j A
lk mk ðm1 ; m2 ; . . . ; mm Þ ¼ exp@  0  where Exkj ; Enkj ; Hekj and fjk are the numerical
2 j¼1  3Enkj 
characteristics (expectation, entropy, hyper-entropy
ð2Þ and order) of index j belonging to grade k; pkjleft and
pkjright represent the widths of the left and right
clouds, respectively; b ¼ 0:01 (Wang et al., 2020a);
Step 1 Determine the classification boundary. The and Lkj is the value of Lkjmin or Lkjmax .
classification grade associated with individual in- Step 3 Construct the connection cloud drop gen-
dex consists of closed interval ½Lkjmin ; Lkjmax . For erator as
the semi-infinite interval ½LKj min ; þ1Þ, the upper 0
 
boundary can be determined as Enkj  N Enkj ; Hekj ð11Þ

   
k0
LK
j max ¼ ExK1
j þ ExK1
j  LK2
j min ð3Þ xkt
j  N Ex k
j ; En j ð12Þ

For the semi-infinite interval ½L1j min ; þ1Þ, the    


upper boundary can be determined as f xkt
j ! li mkj ð13Þ
 
L1j max ¼ Ex2j þ Ex2j  L3j min ð4Þ Repeat t times to generate the normal cloud model.
In Figure 3a, Ex, En and He are related to the
location, width and thickness of the one-dimensional
Step 2 Calculate the numerical characteristics cloud model, respectively. Simultaneously, each
ðEx; En; He; f Þ for ICCM as cloud drop represents a random process of the
ICCM model. By comparison, Figure 3b shows the
Lkjmin þ Lkjmax schematic diagram of the multiple cloud model
Exkj ¼ ð5Þ considering the impact of different indices at the
2
same time, and the vertical axis is the connection
degree of individual samples.
pkj Step 4 Determine the importance of each index
Enkj ¼ ð6Þ using the combination weight method, thus:
3
1 X K  
( rij ¼1 þ li mkj ; ð1  i  n; 1  j
ln K k¼1 ð14Þ
pkjleft ¼ Exkj  Lk1
j min
ðProfit indexÞ  m; 1  k  KÞ
pijright ¼ Lkþ1
j max  Exkj
ð7Þ objective rij
xij ¼ PK ð1  i  n; 1  j  mÞ ð15Þ
k¼1 rij
(
pkjleft ¼ Exkj  Lk1
j min
objective
xij
subjective
xj
ðCost indexÞ xij ¼ Pm ð16Þ
pijright ¼ Lk1
j max  Exkj objective subjective
j¼1 xij xj
ð8Þ
objective subjective
where xij and xij denoted the objective
  and subjective weights, respectively, and xij is the
ln ln94
fjk ¼  k k 
 ð9Þ combination weight of index j associated with sam-
Lj Exj
ln 3En k 
 ple i.
j
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

Figure 3. Schematic diagram of one-dimensional and multiple-dimensional cloud model.

Step 5 Identify the sample grade using the Eu- !


xkk
clidean distance, thus: Precisionk ¼ PK  100% ð23Þ
0   k1 j¼1 xjþ
X m  k  fj
9 m
 ij  Ex j A
lik ¼ exp@ xij   ð17Þ 2  Recall  Precision
2 j¼1  3Enkj 0  F  measure ¼ ð24Þ
Recall þ Precision
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
> disi1 ¼ ðli1  1Þ2 þðli2  0Þ2 þ    þ ðlik  0Þ2 where N is the number of samples, and xkþ and xkj
>
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
<
disi2 ¼ ðli1  0Þ2 þðli2  1Þ2 þ    þ ðlik  0Þ2
are, respectively, the number of samples corre-
ð18Þ sponding to grade k and number of samples that are
>
> ..
>
> . qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
:
classified to class j.
disiK ¼ ðli1  0Þ2 þðli2  0Þ2 þ    þ ðlik  1Þ2

Kih ¼ minðdisih ; 1  h  KÞ ð19Þ


ENGINEERING VALIDATION
where lik represents the composite measurement of OF THE PROPOSED MODEL
sample i, through which the distance disik can be
calculated, and Kih is the evaluation grade of sample Source of Data Used for CBL Evaluation
i.
Step 6: Evaluate model performance using the The updated coal burst database used for CBL
following metrics (Lin et al., 2018): classification by ICCM refers to Pan et al. (2010),
! Meng et al. (2014), Zheng et al. (2014), Cai et al.
1X K
(2016), Hao et al. (2016), Guo et al. (2018), Liu et al.
Accuracy ¼ xkk  100% ð20Þ
N k¼1 (2018), Hu et al. (2019), Yang et al. (2021) and Zhou
et al. (2021b). The CBL assessment was carried out
PK PK using four parameters (GB/T 25,217.2-2010, 2010),
N k¼1 xkk  ðxkþ  xþk Þ
Kappa ¼
2
PK k¼1 ð21Þ namely dynamic failure duration (DT), elastic en-
N  k¼1 ðxkþ  xþk Þ
ergy index (WET), impact energy index (KE) and
! uniaxial compressive strength (RC) (Zhou et al.,
xkk 2018). In Figure 4a, ep and ee represent the perma-
Recallk ¼ PK  100% ð22Þ
j¼1 xþj nent strain and recovery strain after unloading,
respectively. DT is the time from peak loading to
Zhou, Chen, Wei, and Du

Figure 4. Descriptions of parameters for CBL assessment.

complete failure of coal samples under uniaxial CBLs of samples collected from the literature were
loading conditions; it is negatively connected to determined through laboratory tests or the Chinese
CBL. WET is calculated as national standard (GB/T 25,217.2-2010) method. For
the sample grade associated with this method, the
Wst
WET ¼ ð25Þ risk level of the individual index was first deter-
Wsp mined based on its value and classification standard
(Fig. 4b). Then, the results of evaluation of the
where Wst is the elastic energy density and Wsp is the
parameters were considered comprehensively with-
dissipated strain energy, which correspond to the
out the importance of the indicators, through which
area of SAHB and SOHB, respectively. KE is the ratio
the CBL of individual samples can be obtained. The
of the pre-peak area to the post-peak area of the
numerical characteristics of ICCM are shown in
stress–strain curve, thus:
Table 2, including Ex, En, He and f, and its calcu-
uOHGD lation refers to the quantification of index connec-
KE ¼ ð26Þ
Wsf tion, the determination of index importance, the
computation of composite measurement vectors and
where uOHGB and Wsf denote, respectively, the the classification of samples. Figure 5 shows the in-
accumulated strain energy before and after the peak dex connection of the ICCM, which is able to mea-
strength. KE is a measure of the remaining energy sure clearly the degree of membership of each index.
level after complete damage (Ouyang et al., 2020). In Figure 6, a graphical user interface (GUI) was
In Figure 4b, the CBL evaluation indices are divided used to display clearly the risk level and the con-
into three grades, which are presented as ‘‘None,’’ nection relationship, which consists of a background
‘‘Weak’’ and ‘‘Strong,’’ according to Cai et al. graph (Wang et al., 2016), variable inputs, interme-
(2016). diate variables and CBL evaluation results. Evalu-
ation indices were almost benefit indices with DT as
an exception. Then, two-dimensional cloud drops,
Model Calculation i.e., DT vs. WET and KE vs. Rc, are presented in
Figure 7, aiming at considering the impact of dif-
In Table 1, the values of the parameters and ferent indices simultaneously for the CBL evalua-
their risk levels are given to validate the perfor- tion.
mance of the constructed data-driven model. The
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

Table 1. Input variables of samples for the CBL

Samples DT WET KE Rc Actual grade ICCM

1 973 2.7 1.8 5.9 1 2


2 357 2.9 4.6 11.5 2 2
3 49 4.6 6 16.5 3 3
4 291 12.5 3.1 19.3 3 3
5 72 13.2 2.4 19.3 3 3
6 423 5.3 2.6 27.3 3 3
7 112 19.4 4.3 26.9 3 3
8 147 13.3 4.8 30.3 3 3
9 50 8.5 65.9 31.7 3 3
10 459 10 9.9 9.8 2 2
11 23 49 5.1 4.6 3 3
12 42 21.9 5.4 26.6 3 3
13 254 1.6 1.5 12.6 2 2
14 49.667 8.472 65.921 31.661 3 3
15 148.667 2.036 7.368 23.954 3 3
16 321.667 1.246 2.478 13.052 2 2
17 707 12.15 2.8 13.67 2 3
18 208.8 7.32 1.53 13.23 2 2
19 278 10.28 3.39 16.68 3 3
20 278.4 13.36 3.2 20.47 3 3
21 303.4 11.54 2.98 18.18 3 3
22 72 13.35 6.2 19.37 3 3
23 112 19.35 4.26 26.88 3 3
24 613 3.59 3.78 24 2 3
25 39.8 6.49 7.73 24.27 3 3
26 34.4 4.45 12.57 24.35 3 3
27 285 1.91 2.02 10.1 2 2
28 254 1.46 1.59 12.64 2 2
29 31.5 5.6 12 35.4 3 3
30 161.4 3.382 2.253 13.79 2 2
31 133,592 3.93 1.04 4.56 1 2
32 10,702 5.32 12.21 15.17 3 3
33 45 12.3 12.57 18.77 3 3
34 73 2.96 19.37 14.67 3 3
35 158 5.1 6.13 13 3 2
36 60 2.93 13.51 14.32 3 3
37 33 10.28 9.84 12.09 3 3
38 141 5.12 5.6 10.56 3 2
39 82 4.78 10.18 13.9 3 3
40 20 9.43 8.72 16.5 3 3
41 105 7.71 4.21 13.89 3 3
42 54 19.63 1.29 17.25 3 3
43 46.2 14.51 18.78 34.32 3 3
44 12 3.6 2.3 37.7 3 3
45 70 7.83 3.38 20.76 3 3
46 51.25 5.74 1.8 18.5 3 3
47 1414 3.29 2.38 5.26 1 2
48 138.4 0.97 3.54 9.97 2 2
49 52.5 6.62 4.32 9.52 2 3
50 50.35 6.71 4.52 11.78 2 3
51 346 11.3 6.45 15.24 3 3
52 248.57 2.15 1.84 12.03 2 2
53 375 2.1 1.93 11.43 2 2
54 137 5.28 4.15 13.75 2 2
55 258 2.01 2.05 12.49 2 2
56 185 2.78 3.26 13.28 2 2
57 402 5.332 2.632 27.279 3 3
Zhou, Chen, Wei, and Du

Table 1. continued

Samples DT WET KE Rc Actual grade ICCM

58 306 5.91 2.48 8.86 2 2


59 284 3.96 1.84 10.48 2 2
60 351 2.63 1.64 12.98 2 2
61 37 5.33 5.01 15.56 3 3
62 91 4.36 3.89 14.4 2 2
63 144 3.46 3.21 13.35 2 2
64 113 2.78 2.71 12.46 2 2
65 149 2.58 2.37 11.61 2 2
66 362.8 6.02 1.34 8.61 2 2
67 252 3.11 1.93 6.51 2 2
68 100 6.41 2.76 17.15 2 3
69 24 4 2.85 23.85 3 3
70 301 3.84 2.47 10.2 2 2
71 92 4.3 4.53 14.65 2 2
72 300 2.76 2.74 11.76 2 2
73 224 6.44 6.32 18.61 3 3
74 19 7.86 13.667 9.72 3 3
75 41 11.91 11.76 5.489 3 3
76 44.5 4.34 5.99 22.63 3 3
77 139.8 3.19 3.34 13.29 2 2
78 722 1.9 1.54 7.05 1 2
79 760 2.13 1.17 5.92 1 2
80 260 1.88 1.67 12.87 2 2
81 189 6.05 6.49 18.64 3 3
82 260 2.4 0.76 4.27 1 2
83 212 4.34 0.88 7.31 2 2
84 267 12.4 0.87 24.77 3 3
85 65.89 10.42 6.3 28.6 3 3
86 196 1.93 1.078 8.903 2 2
87 288 1.63 2.34 15.37 2 2
88 239 1.4 2.03 11.14 2 2
89 48 8.06 9.4 10.58 3 3
90 34 5.15 6.5 17.35 3 3
91 255 3.4 3.7 11.15 2 2
92 725 1.58 1.4 5.36 1 2
93 3492 6.3 9.75 22.44 3 3
94 18,150 2.5 2.58 7.07 2 2
95 31,325 0.2 1.37 2.04 1 2
96 167 17.603 15.682 22.597 3 3
97 69.2 12.522 35.723 33.907 3 3
98 90 5.27 3.18 18.39 3 3
99 254 3 1.94 7.31 2 2
100 500 1.9 1.63 5.71 2 1
101 750.3 8.19 1.36 9.52 3 3
102 421.8 22.67 1.39 17.4 3 3
103 43.2 14.6 11.8 34 3 3
104 360 2.25 13.84 10 2 2
105 362.8 6.02 1.34 8.61 2 2
106 5.7 9.1 6.48 14.5 3 3
107 38 14.57 17.46 22.59 3 3
108 42 7.39 5.67 20.5 3 3
109 45 5.24 4.96 18.54 3 3
110 49 8.12 10.63 25.9 3 3
111 1149 3.39 3.45 14.38 2 2
112 174 4 1.58 13.17 2 2
113 49 5.09 4.59 23.41 3 3
114 140 6.44 9.51 21 3 3
115 109 5.13 2.19 14.57 2 2
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

Table 1. continued

Samples DT WET KE Rc Actual grade ICCM

116 141 14.74 6.92 14.34 3 3


117 250 3.86 3.15 4.33 2 2
118 141 13.4 2.47 16.55 3 3
119 250 4.69 2.77 9.17 2 2
120 31 16.17 12.17 10.14 3 3
121 203 10.03 4.11 20.37 3 3
122 328 3.4 2.17 5.37 2 2
123 31 2.09 3.25 5.26 2 2
124 438 9.63 5.13 6.82 3 2
125 234 8.15 4.7 11.37 2 2
126 196 1.93 1.08 8.9 2 2
127 70 7.83 3.38 20.8 3 3
128 30 3.38 6.48 13.97 3 3
129 134 3.09 1.42 13.5 2 2
130 160 2.62 1.51 10.36 2 2
131 90 5.27 3.18 18.39 3 3
132 198 2.08 1.82 13.2 2 2
133 356 1.82 1.22 9.64 2 2
134 510 1.25 1.18 7.15 1 1
135 52 7.22 5.15 33.23 3 3
136 359 4.14 1.97 11.93 2 2
137 494 2.88 1.37 10.81 2 2
138 119 4.55 2.95 15.25 2 2
139 58 3.77 2.13 19.75 2 3
140 254 1.59 1.46 12.64 2 2
141 391 4.64 0.9 5.81 2 2
142 90 2.5 2.6 9.51 2 2
143 483 2.08 2.1 1.56 2 2
144 461 2.23 1.45 7.3 2 2
145 537 2.08 2.38 7.79 2 2
146 674 2.23 2.46 8.66 2 2
147 409 2.16 1.39 8.16 2 2

Table 2. Numerical characteristics of the improved multiple dimensional cloud model

Grade Index Ex Enleft Enright fleft fright He

None DT 525.000 158.333 0.000 0.635 0.000 0.01


WET 1.000 0.000 1.333 0.000 1.349 0.01
KE 0.750 0.000 1.417 0.000 1.078 0.01
RC 3.500 0.000 3.500 0.000 1.703 0.01
Weak DT 275.000 91.667 91.667 9.322 9.322 0.01
WET 3.500 1.167 1.167 2.208 2.208 0.01
KE 3.250 1.083 1.083 3.022 3.022 0.01
RC 10.500 3.500 3.500 1.703 1.703 0.01
Strong DT 25.000 0.000 158.333 0.000 0.635 0.01
WET 6.000 1.333 0.000 1.349 0.000 0.01
KE 5.750 1.417 0.000 1.078 0.000 0.01
RC 17.500 3.500 0.000 1.703 0.000 0.01

In this paper, the calculations for sample 2 are according to Eqs. 3–10, as shown in Table 2. The
used as an example. First, the numerical character- expectations (Ex) of DT belonging to different
istics of the aforementioned indices were calculated grades were specified as 525, 275 and 25. The
Zhou, Chen, Wei, and Du

Figure 5. Index connection in the proposed cloud model for CBL.

objective weight of every index was obtained using described by a specified angle. The locations of
the entropy weight theory based on index mea- individual samples shown in the figure are deter-
surement vectors (Eqs. 14 and 15). Then, combined mined by the composite measurement vectors
with the subjective weights {0.4673, 0.16, 0.095, belonging to each grade and specified angle. Simul-
0.2772} raised by Zhou et al. (2021b), the final taneously, a similar distribution characteristic is
weights of sample 2 were presented as {0.56, 0.136, shown in Figure 8b, which determines the location
0.093, 0.211} (after repeating 1000 times) according through the Euclidean distance. As shown in Fig-
to Eq. 16. The connection degrees of sample 2 were ure 8a, the points corresponding to grade 1, i.e.,
{0.082, 0.942, 0.055}, through which the Euclidean ‘‘None,’’ are considerably concentrated and mainly
distance for each grade can be obtained involving distributed on the periphery of polar coordinates in
Eqs. 17 and 18 (i.e., {1.317, 0.114, 1.337}). Then, the Figure 8b. The majority of samples were classified
grade associated with the minimum distance was into risk levels of either ‘‘weak’’ or ‘‘strong,’’ which
considered as the evaluation result (Eq. 19). Thus, are consistent with the outcomes for the confusion
the CBL level of sample 2 was ‘‘weak.’’ matrix.
The calculations for the remaining samples Additionally, the classification performance of
were similar to that of sample 2. In Figure 8a, data the ICCM and other data-driven models were
records for 147 samples were distributed evenly on compared through a confusion matrix and classifi-
the circle of polar coordinates; each sample can be cation metrics (Eqs. 20 and 21), i.e., accuracy and
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

Figure 6. Graphical user interface (GUI) for the CBL evaluation.

Figure 7. Connection degree of the integrated cloud model.

Kappa coefficient. In Figure 9, there are various lighted by different sizes and colors. In Table 3, the
assessment outcomes for different models evenly accuracy and Kappa coefficient of the ICCM are
distributed on the circle and the grades are high- 0.88 and 0.772, respectively. It can be seen from the
Zhou, Chen, Wei, and Du

Figure 8. Composite measurement vectors and the Euclidean distance of the collected database for the CBL assessment.

confusion matrix (Fig. 10) and related metrics that majority of samples belonging to risk levels of
the ICCM performed better than the other models ‘‘weak’’ and ‘‘strong’’ were classified accurately,
for the CBL assessment. Apart from grade 1, the which is beneficial to improving the reliability of
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

Figure 9. Evaluation performance of various models.

underground projects that are designed based on the classify the CBL. In Table 3, the classification
provided evaluation results. Table 4 illustrates the accuracy and Kappa coefficient for the ICCM were
classification performance of models using various 0.88 and 0.772, respectively, which were better than
metrics such as recall, precision and F-measure for the other uncertainty-based models. A user-
(Eqs. 22–24). friendly GUI, which was developed to display vi-
vidly the parameters together with the CBL result,
was able to highlight the parameters needed. More
DISCUSSION importantly, it was able to evaluate the CBL, it
packages quickly the complex process, and it is ea-
In this work, an updated CBL dataset was sier for on-site workers to accept. The classification
compiled to validate the performance of the im- performance of the ICCM on grades ‘‘weak’’ an-
proved cloud model based on set pair analysis and d’’strong’’ was better than that on grade ‘‘none’’
Euclidean distance. Apart from the proposed cloud according to Table 4, which is safer for underground
model, the unascertained measurement theory and a engineering designs. Simultaneously, for given sam-
one-dimensional cloud model were also used to ples, the ICCM was able to evaluate clearly their
Zhou, Chen, Wei, and Du

Table 3. Classification metrics of each model

Classification metrics Unascertained measurement ICCM Normal cloud model

Accuracy (%) 0.84 0.88 0.66


Kappa coefficient 0.722 0.772 0.380

uncertainty of the classification results caused


by the designerÕs judgment. The Euclidean
distance, a reliable cluster methodology, cal-
culates the distance between composite mea-
surement vectors and various grades, making
full use of the existing information to classify
the CBL. In contrast, the maximum measure-
ment principle, which is considered the com-
monly used criterion for previous cloud
models in comprehensive evaluation, was able
to identify the attributes of samples immedi-
ately and directly. For example, in sample 33,
the composite measurement vectors were
{0.014, 0.315, 0.397}. The evaluated CBL was a
grade 3, i.e., ‘‘strong,’’ according to the maxi-
Figure 10. Confusion matrix of the proposed cloud model. mum measurement principle. However, the
information contained in the remaining grades
was ignored when the risk of coal bursts was
CBL without any uncertainty compared to the Chi- evaluated, causing 45% of valuable informa-
nese national standard. The advantages of ICCM tion loss in this system.
were analyzed as follows: (4) Compared to other uncertainty theory-based
algorithms, the ICCM has better performance
(1) Considering the connection degree of interval in uncertainty information treatment. For in-
limitation, the set pair theory was used to de- stance, the unascertained measurement theory
scribe identical–discrepancy–contrary (IDC) has great power on uncertainty information
of each index through the aforementioned interpretation depending on variously devel-
characteristic parameters, which can analyze oped membership functions. However, the
reliably the threshold degree of membership at selection of these functions for special engi-
the junction of adjacent classification intervals. neering projects lies on the hands of re-
(2) In the context of randomness and uncertainty, searchers. Meanwhile, the classification basis,
the ICCM can show the characteristic param- i.e., the credible identification principle, also
eters in the calculation and remove the struggles with the subjective determination as
uncertainty from the evaluation outcomes. The a function of empirical knowledge and rele-
interaction of indices considered in the ICCM vant literature. In comparison, the ICCM is
analyzes index importance before calculating capable of carrying out risk assessment based
the power function, compared to the conven- on a fixed model structure without empirical
tional cloud model. judgment.
(3) For the structure of the ICCM, combination
weight and Euclidean distance were used to
measure index importance and classify the In summary, 147 data records of coal burst cases
CBL. The combination weight can incorporate were employed in the established data-driven mod-
simultaneously the advantage of subjective el, and the results showed that the ICCM has great
and objective weights and it can reduce the potential in coal burst risk assessment over other
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

Table 4. Classification metrics of individual algorithms

Grade Unascertained measurement Improved connection cloud Normal cloud model Support

Recall Precision F-measure Recall Precision F-measure Recall Precision F-measure


(%) (%) (%) (%) (%) (%) (%) (%) (%)

None 0.89 0.38 0.53 0.11 0.50 0.18 0.44 0.80 0.57 9
Weak 0.69 0.98 0.81 0.90 0.85 0.87 0.97 0.58 0.73 68
Strong 0.97 0.87 0.92 0.96 0.92 0.94 0.39 0.96 0.55 70

uncertainty models. Its inspiring accuracy may be sible to construct more data-driven intelligent
attributed to two fundamental aspects. Firstly, the algorithms.
modelÕs structure determines that the ICCM can (2) The proposed ICCM is able to determine
treat uncertainty and randomness information clearly the membership correlation between
effectively and establish the connection between the adjacent intervals and combine the impact of
qualitative concepts and quantitative parameters. various indices through a multiple dimensional
Secondly, the determination of index weight is fea- cloud model. The ICCM had the best classifi-
sible, which incorporates the benefits of different cation performance according to classification
methods considering sample differences. However, metrics, and its accuracy and Kappa coefficient
there is a limitation that should be considered in were 0.88 and 0.772, respectively.
future research, i.e., calculation in the ICCM is re- (3) The subjective weight used in this work was
peated N times; this aims at removing the uncer- based on previous research. The way to
tainty of generated random numbers, but the quantify the index importance using other
amount of calculation will likely as the index system highly efficient methods and to combine these
becomes complex. Thus, the optimum value of N weights into ICCM has not yet been explored.
should be investigated scientifically by integrating Furthermore, the N times of calculation for the
other methods. current cloud models need to be investigated
further to balance the relationship between the
reliability of outcomes and time complexity.
CONCLUSIONS

In this work, ICCM was developed based on the


cloud model theory and the set pair analysis to
ACKNOWLEDGMENTS
evaluate CBL based on an updated coal burst da-
taset. The ICCM can be described by four numerical
This research was funded by the National Sci-
characteristics, i.e., Ex, En, He and f. Meanwhile,
ence Foundation of China (Nos. 42177164 and
entropy weight and AHP were integrated to calcu-
41807259), the Innovation-Driven Project of Central
late the importance of each index based on index
South University (No. 2020CX040), the Natural
measurement vectors. Euclidean distance was used
Science Foundation of Hunan Province
to classify the level of CBL, and it was able to pre-
(2020JJ5743) and the Accurate Delay Rock Break-
vent loss of valuable information effectively. Ulti-
ing Mechanism And Key Technology Innovation
mately, CBL evaluation was carried out vividly
Team (No. 2020D14043).
through a user-friendly GUI that focuses on the
specified parameters. Our conclusions are as follows
(1) Due to uncertainty and complexity of geolog- DATA AVAILABILITY
ical conditions, there is a requirement for an
evaluation methodology that is able to elimi- All data generated or analyzed during this study
nate fuzziness from results. In this study, 147 are included in this published article.
data records of coal burst cases made it pos-
Zhou, Chen, Wei, and Du

CODE AVAILABILITY GB/T 25217.2-2010 (2010). Classification and laboratory test


method on bursting liability of coal. Beijing: Standards Press
of China.
The used codes are available upon request. Guo, W. Y., Tan, Y. L., Yang, Z. L., Zhao, T. B., & Hu, S. C.
(2018). Effect of saturation time on the coal burst liability
DECLARATIONS indexes and its application for rock burst mitigation.
Geotechnical and Geological Engineering, 36(1), 589–597.
Han, H., Armaghani, D. J., Tarinejad, R., Zhou, J., & Tahir, M.
Conflict of Interest The authors declare no con- M. (2020). Random forest and bayesian network techniques
flict of interest. for probabilistic prediction of flyrock induced by blasting in
quarry sites. Natural Resources Research, 29(2), 655–667.
Hao, Y. X. (2016). Study on rockburst of Wudong near-vertical
Coal Seams and impact protection with constant resistance
and large deformation supporting. China university of mining
and technology.
REFERENCES Hu, S., & Zhang, H. (2019). Research on rockburst risk assess-
ment and prevention techniques in Changcun Coal Mine.
Aghababaei, S., Saeedi, G., & Jalalifar, H. (2016). Risk analysis China Energy and Environmental Protection., 41(6), 15–18.
and prediction of floor failure mechanisms at longwall face in Li, C., Zhou, J., Armaghani, D. J., Cao, W., & Yagiz, S. (2021a).
parvadeh-I coal mine using rock engineering system (RES). Stochastic assessment of hard rock pillar stability based on
Rock Mechanics and Rock Engineering., 49(5), 1889–1901. the geological strength index system. Geomechanics and
Cai, W., Dou, L., Si, G., Cao, A., He, J., & Liu, S. (2016). A Geophysics for Geo-Energy and Geo-Resources, 7(2), 1–24.
principal component analysis fuzzy comprehensive evalua- Li, C., Zhou, J., Armaghani, D. J., & Li, X. (2021b). Stability
tion model for coal burst liability assessment. International analysis of underground mine hard rock pillars via combi-
Journal of Rock Mechanics and Mining Sciences., 100(81), nation of finite difference methods, neural networks, and
62–69. Monte Carlo simulation techniques. Underground Space,
Cai, W., Dou, L., Zhang, M., Cao, W., Shi, J. Q., & Feng, L. 6(4), 379–395.
(2018). A fuzzy comprehensive evaluation methodology for Li, D. Y., Liu, C. Y., & Liu, L. Y. (2004). Study on the universality
rock burst forecasting using microseismic monitoring. Tun- of the normal cloud model. Engineering Science., 6(8), 28–34.
nelling and Underground Space Technology., 80, 232–245. Li, E., Yang, F., Ren, M., Zhang, X., Zhou, J., & Khandelwal, M.
Chen, C., Zhou, J., Zhou, T., & Yong, W. (2021a). Evaluation of (2021c). Prediction of blasting mean fragment size using
vertical shaft stability in underground mines: Comparison of support vector regression combined with five optimization
three weight methods with uncertainty theory. Natural Ha- algorithms. Journal of Rock Mechanics and Geotechnical
zards., 109, 1457–1479. Engineering, 13(6), 1380–1397.
Chen, F., Cao, A., Liang, Z., & Liu, Y. (2021b). A coal burst risk Li, E., Zhou, J., Shi, X., Armaghani, D. J., Yu, Z., Chen, X., &
assessment model of seismic events based on multiple seismic Huang, P. (2021d). Developing a hybrid model of salp swarm
source parameters: A case study of the Huating Coal Mine, algorithm-based support vector machine to predict the
Gansu Province, China. Natural Resources Research, 30, strength of fiber-reinforced cemented paste backfill. Engi-
4515–4532. neering with Computers, 37(4), 3519–3540.
Chen, F., Cao, A., Dou, L., & Jing, G. (2019). A quantitative Li, J., Li, B., Cheng, Q., & Gao, Z. (2021e). Evolution of Ani-
evaluation method of coal burst hazard based on zone divi- sotropic Coal Permeability Under the Effect of Heteroge-
sion and an analytic hierarchy process: A case study on neous Deformation of Fractures. Natural Resources
Yanbei coal mine, Gansu Province. China. Geosciences Research, 1–20.
Journal, 23(5), 833–848. Li, N., Sun, W., Huang, B., Chen, D., Zhang, S., & Yan, M.
Ding, Z., Nguyen, H., Bui, X. N., Zhou, J., & Moayedi, H. (2020). (2021f). Acoustic emission source location monitoring of
Computational intelligence model for estimating intensity of laboratory-scale hydraulic fracturing of coal under true tri-
blast-induced ground vibration in a mine based on imperialist axial stress. Natural Resources Research, 30(3), 2297–2315.
competitive and extreme gradient boosting algorithms. Nat- Li, X. B., Zhou, J., Wang, S. F., & Liu, B. (2017). Review and
ural Resources Research, 29(2), 751–769. practice of deep mining for solid mineral resources. Chinese J
Dai, Y., Khandelwal, M., Qiu, Y., Zhou, J., Monjezi, M., & Yang, Nonferrous Metals., 27(7), 1236–1262.
P. (2022). A hybrid metaheuristic approach using random Qi, Q. X., Li, Y. Z., Zhao, S. K., Zhang, N. B., Zheng, W. Y., Li,
forest and particle swarm optimization to study and evaluate H. T., & Li, H. Y. (2019). Seventy years development of coal
backbreak in open-pit blasting. Neural Computing and mine rockburst in China: Establishment and consideration of
Applications. https://doi.org/10.1007/s00521-021-06776-z. theory and technology system. Coal Science and Technology.,
Du, K., Li, X. F., Su, R., Tao, M., Lv, S. Z., Luo, J., & Zhou, J. 47(9), 1–40.
(2022). Shape ratio effects on the mechanical characteristics Lin, Y., Zhou, K., & Li, J. (2018). Application of cloud model in
of rectangular prism rocks and isolated pillars under uniaxial rock burst prediction and performance comparison with three
compression. International Journal of Mining Science and machine learning algorithms. IEEE Access., 6, 30958–30968.
Technology, 1–16. https://doi.org/10.1016/j.ijmst.2022.01.004. Liu, Z., Cao, A., Guo, X., & Li, J. (2018). Deep-hole water
Du, K., Li, X. F., Yang, C. Z., Zhou, J., Chen, S. J., & Manoj, K. injection technology of strong impact tendency coal seam—a
(2020). Experimental investigations on mechanical perfor- case study in Tangkou coal mine. Arabian Journal of Geo-
mance of rocks under fatigue loads and biaxial confinements. sciences., 11(2), 1–9.
Journal of Central South University, 27(10), 2985–2998. Lu, C. P., Liu, G. J., Liu, Y., Zhang, N., Xue, J. H., & Zhang, L.
Du, K., Sun, Y., Zhou, J., Wang, S. F., Tao, M., Yang, C., & (2015). Microseismic multi-parameter characteristics of
Khandelwal, M. (2021). Low amplitude fatigue performance rockburst hazard induced by hard roof fall and high stress
of sandstone, marble, and granite under high static stress. concentration. International Journal of Rock Mechanics and
Geomechanics and Geophysics for Geo-Energy and Geo- Mining Sciences., 76, 18–32.
Resources, 7(3), 1–21. Makówka, J. (2016). Rock burst: diagnosing and mitigating. Polish
experience for Australian conditions. In: Focused Interna-
An Improved Connection Cloud Model of an Updated Database: A Multicriteria

tional Workshop on Coal Burst Experience and Research Xu, X. F., 2011. Research of mechanism and controlling tech-
Direction. UNSW Australia, Sydney. nology of floor burst in coal seam roadway. Doctoral dis-
Mark, C. (2014). Coal bursts in the deep longwall mines of the sertation. Xuzhou, Jiangsu: China University of Mining and
United States. In: Proceedings 3rd Australasian Ground Technology.
Control in Mining Conference. Victoria Australasian Insti- Yang, L., Wang, X. Q., & Li, J. Z. (2021). Energy evolution and
tute of Mining and Metallurgy, Carlton, pp. 33–39. damage characteristics of coal with different bursting liability
Meng, H. D., Hou, X. Z., Li, H. X., & Zhao, Q. (2014). Moni- under uniaxial compression. Coal Science and Technology.,
toring and prevention research of working face rock burst in 49(6), 111–118.
deep coal well with large dip angle. Coal Technology., 33(10), Yu, Z., Shi, X., Miao, X., Zhou, J., Khandelwal, M., Chen, X., &
63–65. Qiu, Y. (2021). Intelligent modeling of blast-induced rock
Ortlepp, W. D. (2005). RaSiM comes of age-a review of the movement prediction using dimensional analysis and opti-
contribution to the understanding and control of mine rock- mized artificial neural network technique. International
bursts. In Proceedings of the sixth international symposium on Journal of Rock Mechanics and Mining Sciences., 143,
rockburst and seismicity in mines, Perth, Western Australia 104794.
(pp. 9–11). Zhang, C., Canbulat, I., Hebblewhite, B., & Ward, C. R. (2017).
Ouyang, Z., Zhang, G., Li, Q., Lai, X., Qin, H., Zhao, X., & Zhou, Assessing coal burst phenomena in mining and insights into
X. (2020). Study on the rock burst tendentiousness of coal directions for future research. International Journal of Coal
under different gas pressures. Arabian Journal of Geo- Geology., 179, 28–44.
sciences, 13(1), 1–13. Zhang, L., Huang, M., Xue, J., Li, M., & Li, J. (2021). Repetitive
Pan, J., Mao, D., Lan, H., Wang, S., & Qi, Q. (2013). Study status mining stress and pore pressure effects on permeability and
and prospects of mine pressure bumping control technology pore pressure sensitivity of bituminous coal. Natural Re-
in China. Coal Sci. Technol., 41(6), 21–25. sources Research, 30, 4457–4476.
Pan, Y. S., Geng, L., & Li, Z. H. (2010). Research on evaluation Zhang, L., Wu, X., Chen, Q., Skibniewski, M. J., & Zhong, J.
indices for impact tendency and danger of coal seam. Journal (2015). Developing a cloud model based risk assessment
of China Coal Society., 35(12), 1975–1978. methodology for tunnel-induced damage to existing pipe-
Sabapathy, R., Paul, P. S., & Mandal, P. K. (2019). Evaluation of lines. Stochastic Environmental Research and Risk Assess-
bump-proneness of underground coal mines using burst en- ment, 29(2), 513–526.
ergy coefficient. Arabian Journal of Geosciences., 12(18), 1– Zheng, S. D. (2014). Exploration on comprehensive prevention
16. and control technology of rock burst in Gengcun. Coal Mine.,
Sahoo, M., Kumar, A., Chalavadi, G., Gupta, A., & Singh, R. K. 3, 211–212.
(2021). Enhancement of dry beneficiation of coal by air dis- Zhou, J., Chen, C., Armaghani, D. J., & Ma, S. (2020a). Devel-
tributor design modification in an air-pulsated stratifier. oping a hybrid model of information entropy and unascer-
Natural Resources Research, 2, 1–16. tained measurement theory for evaluation of the
Tang, C., Yao, Q., Xu, Q., Shan, C., Xu, J., Han, H., & Guo, H. excavatability in rock mass. Engineering with Computers, 1–
(2021). Mechanical failure modes and fractal characteristics 24. https://doi.org/10.1007/s00366-020-01053-4.
of coal samples under repeated drying-saturation conditions. Zhou, J., Chen, C., Du, K., Armaghani, D. J., & Li, C. (2020b). A
Natural Resources Research, 2, 1–18. new hybrid model of information entropy and unascertained
Wang, G. F., Gong, S. Y., Li, Z. L., Dou, L. M., Cai, W., & Mao, measurement with different membership functions for eval-
Y. (2016). Evolution of stress concentration and energy re- uating destressability in burst-prone underground mines.
lease before rock bursts: Two case studies from Xingan coal Engineering with Computers, 1–19. https://doi.org/10.1007/s0
mine, Hegang, China. Rock Mechanics and Rock Engineer- 0366-020-01151-3.
ing., 49(8), 3393–3401. Zhou, J., Chen, C., Khandelwal, M., Tao, M., & Li, C. (2021a).
Wang, M., Liu, Q., Wang, X., Shen, F., & Jin, J. (2020a). Pre- Novel approach to evaluate rock mass fragmentation in block
diction of rockburst based on multidimensional connection caving using unascertained measurement model and infor-
cloud model and set pair analysis. International Journal of mation entropy with flexible credible identification criterion.
Geomechanics., 20(1), 04019147. Engineering with Computers, 1–21. https://doi.org/10.1007/s0
Wang, S. M., Zhou, J., Li, C. Q., Armaghani, D. J., Li, X. B., & 0366-020-01230-5.
Mitri, H. S. (2021). Rockburst prediction in hard rock mines Zhou, J., Chen, C., Wang, M., & Khandelwal, M. (2021b).
developing bagging and boosting tree-based ensemble tech- Proposing a novel comprehensive evaluation model for the
niques. Journal of Central South University, 28(2), 527–542. coal burst liability in underground coal mines considering
Wang, X., Shi, K., Shi, Q., Dong, H., & Chen, M. (2020b). A uncertainty factors. International Journal of Mining Science
normal cloud model-based method for risk assessment of and Technology., 31(5), 799–812.
water inrush and its application in a super-long tunnel con- Zhou, J., Dai, Y., Khandelwal, M., Monjezi, M., Yu, Z., & Qiu, Y.
structed by a tunnel boring machine in the arid area of (2021c). Performance of hybrid SCA-RF and HHO-RF
northwest China. Water, 12(3), 644. models for predicting backbreak in open-pit mine blasting
Wang, Y., Yang, W., Li, M., & Liu, X. (2012). Risk assessment of operations. Natural Resources Research, 30, 4753–4771.
floor water inrush in coal mines based on secondary fuzzy Zhou, J., Huang, S., Wang, M., & Qiu, Y. (2021d). Performance
comprehensive evaluation. International Journal of Rock evaluation of hybrid GA–SVM and GWO–SVM models to
Mechanics and Mining Sciences., 52, 50–55. predict earthquake-induced liquefaction potential of soil: A
Wei, C., Zhang, C., Canbulat, I., Cao, A., & Dou, L. (2018). multi-dataset investigation. Engineering with Computers, 1–
Evaluation of current coal burst control techniques and 19. https://doi.org/10.1007/s00366-021-01418-3.
development of a coal burst management framework. Tun- Zhou, J., Li, X., & Mitri, H. S. (2018). Evaluation method of
nelling and Underground Space Technology, 81, 129–143. rockburst: State-of-the-art literature review. Tunnelling and
Xie, C., Nguyen, H., Bui, X. N., Nguyen, V. T., & Zhou, J. (2021). Underground Space Technology, 81, 632–659.
Predicting roof displacement of roadways in underground Zhou, J., Shen, X., Qiu, Y., Li, E., Rao, D., & Shi, X. (2021e).
coal mines using adaptive neuro-fuzzy inference system Improving the efficiency of microseismic source locating
optimized by various physics-based optimization algorithms. using a heuristic algorithm-based virtual field optimization
Journal of Rock Mechanics and Geotechnical Engineering, method. Geomechanics and Geophysics for Geo-Energy and
13(6), 1452–1465. Geo-Resources., 7(3), 89.
Zhou, Chen, Wei, and Du

Zhou, J., Qiu, Y., Khandelwal, M., Zhu, S., & Zhang, X. (2021f). Zhou, J., Zhu, S., Qiu, Y., Armaghani, D. J., Zhou, A., & Yong,
Developing a hybrid model of Jaya algorithm-based extreme W. (2022b). Predicting tunnel squeezing using support vector
gradient boosting machine to estimate blast-induced ground machine optimized by whale optimization algorithm. Acta
vibrations. International Journal of Rock Mechanics and Geotechnica. https://doi.org/10.1007/s11440-022-01450-7.
Mining Sciences, 145, 104856. Zhou, K. P., Lin, Y., Deng, H. W., Li, J. L., & Liu, C. J. (2016).
Zhou, J., Huang, S., Zhou, T., et al. (2022a). Employing a genetic Prediction of rock burst classification using cloud model with
algorithm and grey wolf optimizer for optimizing RF models entropy weight. Transactions of Nonferrous Metals Society of
to evaluate soil liquefaction potential. Artificial Intelligence China, 26(7), 1995–2002.
Review. https://doi.org/10.1007/s10462-022-10140-5.

You might also like