Ain Shams Engineering Journal xxx (xxxx) xxx

Contents lists available at ScienceDirect

Ain Shams Engineering Journal

journal homepage: www.sciencedirect.com

Engineering Physics and Mathematics

A novel hyperbolic time-delayed grey model with Grasshopper

Optimization Algorithm and its applications
Xiwang Xiang a, Xin Ma a,⇑, Yizhu Fang b, Wenqing Wu a, Gaoxun Zhang a
a
School of Science, Southwest University of Science and Technology, Mianyang 621010, China
b
Exploration and Development Research Institute, Petrochina Southwest Oil & Gasfield Company, Chengdu 610051, China

a r t i c l e i n f o a b s t r a c t

Article history: Discharge for wastewater treatment plays a key role in improving the water quality, thereby guarantee-
Received 27 January 2020 ing living quality of citizens. With high-speed economics growth and economics reforming, total amount
Revised 17 July 2020 of China’s discharge of wastewater treatment is sharing high uncertainty, leading to many difficulties in
Accepted 25 July 2020
accurate forecasts of discharge of wastewater treatment. Based on grey system theory, the hyperbolic
Available online xxxx
time-delayed term is introduced in this paper to develop a novel forecasting model in order to deal with
uncertainties of China’s sewage discharge forecasting. The key nonlinear parameter of the proposed
Keywords:
model is determined by the Grasshopper Optimization Algorithm. A series of practical numerical cases
Discharge of sewage
Ecological civilization construction
prove that the proposed model is reliable in comparison with six existing models are used for compari-
Hyperbolic time-delayed grey model son. Then we apply it to predict the behavior of sewage discharge in China, those results against demon-
Grasshopper optimization algorithm strating the model our proposed has more satisfactory prediction precision.
Prediction precision Ó 2020 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams Uni-
versity. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/
by-nc-nd/4.0/).

1. Introduction work. To sum up, accurate prediction of sewage discharge by scien-

tific means is not only directly related to the construction and
Water is the source of life and the foundation of human survival development of advanced sewage treatment industry but also
and development. The situation of water shortage in China has has a positive impact on the social progress, economic develop-
been severe for a long time. Due to the rapid process of industrial- ment, and ecological environment construction of the country.
ization and urbanization, China is facing a growing demand for In recent decades, a large number of sewage prediction models
energy and a large number of domestic sewage and industrial based on different mathematical models or mechanisms have been
wastewater are being discharged [1,2]. The increasingly tense developed, and some satisfactory results have been obtained.
situation between the scarcity of water resources and worsening Suchetana et al. [4,5] applied the regression model to study the
environmental pollution has severely affected people’s living envi- change of pollutant concentration in wastewater and achieved
ronment, threatened people’s health, and become an important some results. Regression models are also used to predict wastew-
factor restricting China’s sustainable economic development, and ater discharge and some of its essential water quality parameters
also promoted sewage treatment to become one of the critical [6,7]. Li et al. [8] utilized four different curve formulas to fit the dis-
infrastructure projects for sustainable urban development [3]. charge of restaurant wastewater in Beijing from 2004 to 2007, and
The prediction of sewage discharge is a prerequisite for the plan- the discharge of restaurant wastewater from 2008 to 2012 was
ning, design, operation and management of drainage systems, predicted to control environmental pollution in Beijing. Also, Time
especially in the optimal design of the urban drainage pipe net- series models became popular and were widely used in environ-
mental forecasting due to its simple structure and fast conver-
⇑ Corresponding author. gence. Man et al. [9] designed a hybrid algorithm based on the
E-mail address: [email protected] (X. Ma). auto-regressive moving average (ARMA) and the vector auto-
Peer review under responsibility of Ain Shams University. regression (VAR) to predict the sewage inflow and chemical oxygen
demand (COD) in sewage. Liu and his collaborators [10,11] succes-
sively employed the auto-regressive (AR) models and the auto-
regressive with exogenous inputs (ARX) models to predict the flow
Production and hosting by Elsevier of sewage to help them better control the chemical dosing rate in

https://doi.org/10.1016/j.asej.2020.07.019
2090-4479/Ó 2020 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article as: X. Xiang, X. Ma, Y. Fang et al., A novel hyperbolic time-delayed grey model with Grasshopper Optimization Algorithm and its
applications, Ain Shams Engineering Journal, https://doi.org/10.1016/j.asej.2020.07.019
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

sewer networks. In the past few decades, artificial intelligence (AI) [40,41] introduced the time delay term into the polynomial grey
algorithms have also been paid increasing attention in the sewage model and fractional grey model, and the results show those are
treatment industry. Wei et al. [12] proposed a dynamic neural net- quite effective in improving the prediction accuracy of the grey
work model with an online corrector to predict the influent flow of model.
wastewater treatment plants, and they found that this method Nowadays, most of the time delay term in the existing time-
could effectively solve the time lag problem and improve the accu- delayed grey models are considered fixed, limiting the time-
racy for longer time horizons. Mozafar et al. [13] integrated two delayed grey model’s performance. Due to this reason, we were
intelligent algorithms to optimize the fuzzy inference system motivated to investigate a new time-delayed grey model with
(FIS) model, and the proposed model successfully provide out- dynamic features to reverse the awkward situation. Therefore, a
standing performance for predicting six well-known influent novel hyperbolic time-delayed grey model with an adjustable
chemical and biological characteristics. Zhou [14] proposed two parameter is proposed in this paper, denoted as the HTGM (1,1)
artificial neural networks (ANN) models improved by Multivariate model, which could maturely interpret the nonlinear disturbances.
Bayesian Uncertainty Processor (MBUP) to predict water quality, Then six practical numerical cases are used to verify HTGM (1,1)
and practical cases verified the reliability of the method. Recently, model is reliable, and compared it with six classical grey models
the ANN and other AI algorithms have always been applied to during verification. Finally, HTGM (1,1) model is applied to predict
predict some of the key data in the pollutant treatment process the behavior about sewage discharge in China.
[15–17]. In general, every single model has its own inevitable lim- The remainder of this paper is organized as follows: a brief
itations so that the hybrid prediction models are easily considered. overview of the classical grey GM (1,1) model is introduced in Sec-
Du and his collaborators [18–21] confirmed this claim and illus- tion 2. The hyperbolic grey model HTGM (1,1) and its modelling
trated its effectiveness through practical application cases. More- mechanism are explained in detail in Section 3, then the prepara-
over, some researchers analysed the structure of the target to tion of nonlinearly constrained optimization problem with its solu-
describe its behaviour rule, which also is a novel idea worth con- tion by the Grasshopper Optimization Algorithm is provided in
sidering [22–24]. However, it would be captious to collect large Section 4. The validation cases of the HTGM (1,1) model are drawn
amounts of reliable data about the sewage discharge in China. in Section 5. In Section 6, the sewage discharge in China is studied
The reason why that is the government’s definition of relevant as an application and discussed, which includes the domestic sew-
indicators is vague, and the accounting methods are different in age discharge and the industrial wastewater discharge in China. At
the early stage. And the iterative updating of monitoring equip- the end of this paper, we summarize the whole research work and
ment also could make the monitoring data inaccurate or question- draw a conclusion in Section 7.
able. To hack these challenges, the grey model is another
alternative option, which is known for its excellent ability to solve 2. Overview of classical grey GM (1,1) model
those problems with small samples and poor information.
Grey model is one of the most conspicuous achievements of Accumulated generation operation is a marvelous achievement
grey system theory [25], who is always popular when dealing with that arises from the grey system theory, which laid the foundation
uncertain systems, even in situations where there are fewer mod- for the development of the grey prediction model. In this section,
elling data. For decades, many researchers devoted themselves to we introduce the define of accumulated generation operation and
the grey model and applied it successfully in many fields. To the theory about the classical grey model.
enhance the precision of the grey model, Xu et al. [26] investigated

In general, for an original data sequence X ð0Þ ¼ xð0Þ ð1Þ;
the relationship between the initial value and time input, thus pro-
xð0Þ ð2Þ;:::;xð0Þ ðnÞÞ, The accumulated generation operatorðAGOÞ of X ð0Þ
posed a novel grey model with optimized the initial value. Zeng
can be defined as follows.
et al. [27] designed a self-adaptive intelligence grey model to
increase the applicability of the grey model. Ding et al. [28] and

Zhao et al. [29] studied a grey prediction model with a rolling mod- Definition 2.1. Let the sequence X ð1Þ ¼ xð1Þ ð1Þ; xð1Þ ð2Þ; :::; xð1Þ ðnÞ
elling mechanism. Yu et al. [30] optimized the background value of be the accumulated generating operatorð AGOÞ of X ð0Þ , where
the model to improve the performance of the grey model. He et al.
X
k
[31] utilize grey models to construct a hybrid model and received xð1Þ ðkÞ ¼ xð0Þ ðiÞ; k ¼ 1; 2; :::; n: ð1Þ
satisfactory results. Wu et al. [32] thought the key to boosting the i¼1


prediction accuracy lies in the priority of new information. All Similarly, we can obtain the restored value X ð0Þ ¼ xð0Þ ð1Þ;xð0Þ ð2Þ;:::;
these works promote the development of the grey model to some xð0Þ ðnÞÞ, which is called the inverse accumulated generating opera-
extent. The classic grey model didn’t fully consider the influence of
torðIAGOÞ of X ð1Þ , where
external factors on the development trend of the system. Some
researchers investigated the grey models with multiple variables
xð0Þ ðkÞ ¼ xð1Þ ðkÞ  xð1Þ ðk  1Þ; k ¼ 1; 2; :::; n: ð2Þ
to remedy this defect, and the effectiveness of the multivariable
grey model was demonstrated by practical cases. Then Zeng et al.
[33] further generalized and optimized the multivariable grey
model. Thereafter, Ma [34] thought it is promising to introduce

Definition 2.2. Let the X ð0Þ ¼ xð0Þ ð1Þ; xð0Þ ð2Þ;    ;
sequence
machine learning to the grey model. The fractional grey model

xð0Þ ðnÞÞbe the original sequence, and X ð1Þ ¼ xð1Þ ð1Þ; xð1Þ ð2Þ;    ;
was the focus of research in recent years and attracted many schol-
ars to pay attention [35]. Ma et al. [36] applied the discrete mod- xð1Þ ðnÞÞbe the accumulated generation operator of X ð0Þ . The back-


elling technique to the fractional multiple grey model. Wu et al. ground value sequence Z ð1Þ ¼ zð1Þ ð2Þ; zð1Þ ð3Þ;    ; zð1Þ ðnÞ generated
[37] took the Bernoulli equation into the fractional grey model by background value of X ð1Þ , where
and achieved good results. Recently, Ma et al. [38] provided a con-
1
ð1Þ 
formable fractional grey model, which provided a new idea for zð1Þ ðkÞ ¼ x ðkÞ þ xð1Þ ðk  1Þ ; k ¼ 2; 3;    ; n: ð3Þ
modelling. The time delay term was regarded as an effective tech- 2
The equation
nique to improve the precision of the grey model. Hu et al. [39]
constructed the multivariate time-delayed grey model and solved
its parameters by the intelligent algorithm. Later, Ma et al.
xð0Þ ðkÞ þ azð1Þ ðkÞ ¼ b ð4Þ

2
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

is the basic form of the GM (1,1) model, where a is the development In particular, the hyperbolic time-delayed function
P½t
s¼1 tanhðxsÞ ¼ 0 when x ¼ 0, which corresponds to a horizontal
coefficient, bis the grey action quantity. The parameters set
T
u ¼ ½a; b can be estimated by the least squares method, and satisfies line in Fig. 1 independent of time t. The whitening differential
1 equation of the HTGM (1,1) can be expressed as:
u ¼ BT B BT Y ð5Þ
dxð1Þ ðtÞ
þ axð1Þ ðt Þ ¼ c: ð12Þ
where dt
2 3
2  3 Thus, the classical grey model GM (1,1) [25] is obtained. And
xð0Þ ð2Þ 12 xð1Þ ð2Þ þ xð1Þ ð1Þ 1
6
 7 when x ¼ 1, the whitening differential equation of the HTGM
6 xð0Þ ð3Þ 7 6  12 xð1Þ ð3Þ þ xð1Þ ð2Þ 17
6 7 6 7 (1,1) can be expressed as:
Y¼6 7
6 .. 7; B ¼ 6 .. .. 7 : ð6Þ
4 . 5 6 . . 7
4 5 dxð1Þ ðtÞ

 þ axð1Þ ðt Þ ¼ bðtanh1 þ tanh2 þ    þ tanh½tÞ þ c: ð13Þ
xð0Þ ðnÞ  12 xð1Þ ðnÞ þ xð1Þ ðn  1Þ 1 dt
For convenience, the Eq. (13) can also be written as:
ð1Þ
Definition 2.3. Assume that all the parameters are the same as dx ðtÞ
þ axð1Þ ðt Þ ¼ bf ðtÞ þ C; ð14Þ
those defined previously, dt

dxð1Þ ðtÞ where f ðt Þ ¼ tanh½t ; C ¼ bðtanh1 þ tanh2 þ    þ tanhð½t   1ÞÞ þ c.

þ axð1Þ ðtÞ ¼ b ð7Þ Then the HTGM (1,1) model can be approximately degraded to
dt
the HTGM without time-delayed effect, which can also be named
is called the whitening differential equation of the GM (1,1) model. as the standard hyperbolic grey model.
Combining the whitening differential and initial condition Integrating both sides of the whitening differential equation Eq.
(xð0Þ ð1Þ ¼ xð1Þ ð1Þ), the time response sequence can be easily solved (10), one obtains:
as:
Z Z Z X
½t 
  k k k
b b xð 1 Þ ð t Þ þ a xð1Þ ðtÞdt ¼ b tanhðxsÞdt þ c; ð15Þ
^ð1Þ
x ðkÞ ¼ xð0Þ ð1Þ  eaðk1Þ þ ; k ¼ 1; 2;    ; n: ð8Þ
a a k1 k1 k1 s¼1
Later, applying the inverse accumulated generation operation to simplified Eq. (15) with trapezoidal integral formula we can obtain:
Eq. (8), we have the following restored values as: !
b Xk X
k1
  ð0Þ
x ðkÞ þ az ðkÞ ¼ ð1 Þ
tanhðxsÞ þ tanhðxsÞ þ c; ð16Þ
b 2 s¼1
^xð0Þ ðkÞ ¼ ð1 - ea Þ xð0Þ ð1Þ  eaðk1Þ ; k ¼ 2; 3; . . . n: ð9Þ s¼1
a
where zð1Þ ðkÞ is the background value of the HTGM (1,1) model,
3. The hyperbolic time-delayed grey model which has the same definition as Definition 2.2.

Although most of the existing time-delayed grey models boost 3.2. The linear parameter estimation of the HTGM (1,1) model
the capability of the grey model to a certain extent, they do not
always provide a satisfactory result for all problems. The main rea- It would be a challenge to solve the equation system Eq. (10)
son for this is the time delay term is fixed. In our work, a novel directly, but it is worth noting Eq. (16) would degenerate into a
hyperbolic time-delayed grey model with an adjustable parameter compact linear system once the parameter x is controlled. Then
is proposed. Obviously, the proposed model is more general and we can estimate the linear parameter a, b and c based on the least
more flexible in solving problems with disturbances. More critical square rule.
details about our proposed model are presented in this section. In accordance with the least square rule, the linear parameter
 1  
T
3.1. The representation of the HTGM (1,1) model set can be written as u ¼ ½a; b; c ¼ BT B BT Y , where
2 P  3
i¼1 tanhðxsÞ þ tanhðxÞ
2
2 3 zð1Þ ð2Þ 1
1
xð0Þ ð2Þ 6 P
2
7 
6 xð0Þ ð3Þ 7 6 ð1Þ P2 7
i¼1 tanhðxsÞ þ i¼1 tanhðxsÞ
6 7 6 z ð3Þ 1 3
17

 6 7
Y¼6 7 2
Definition 3.1. Assume X ð1Þ ¼ xð1Þ ð1Þ; xð1Þ ð2Þ;    ; xð1Þ ðnÞ has the 6 .. 7; B¼6
6 .. ..
7:
.. 7
4 . 5 6
same definition as Definition 2.3, the whitening differential 4 . P .  .75
xð0Þ ðnÞ Pn1
i¼1 tanhðxsÞ þ i¼1 tanhðxsÞ
n
equation zð1Þ ðnÞ 1
2
1
ð17Þ
dxð1Þ ðtÞ X½t 
þ axð1Þ ðtÞ ¼ b tanhðxsÞ þ c ð10Þ
dt s¼1
3.3. The solution of the HTGM (1,1) model
is named as hyperbolic time-delayed grey model, denoted as the
HTGM (1,1), where hyperbolic time-delayed term be defined as Retrospect to the whitening differential equation Eq. (10), con-
P½t
s¼1 tanhðxsÞ. ½t  is the smallest integer not less than t , also known
sidering the homogeneous differential equation corresponding to it
as the ceiling function and it is defined as follows: we have:
dxð1Þ ðtÞ
½t ¼ minfn 2 Zjt 6 ng: ð11Þ þ axð1Þ ðt Þ ¼ 0; ð18Þ
dt
Moreover, the relationship between the hyperbolic time- the solution of Eq. (18) can be easily solved as:
P½t
delayed function s¼1 tanhðxsÞ and nonlinear parameters x is xð1Þ ðt Þ ¼ Cxð1Þ ð1Þeaðt1Þ : ð19Þ
illustrated in Fig. 1. It can be clearly seen that the time path t exerts
different nonlinear effect on the hyperbolic time-delayed function On the basis of the primary theory about infinitesimal calculus,
with different nonlinear parameter x. we assume the form of the solution about Eq. (19) is
3
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

(a) (b)
P½t
Fig. 1. Disturbance of s¼1 tanhðxsÞ by t: (a) 0 6 x < 1, (b) 1 < x 6 0.

xð1Þ ðtÞ ¼ C ðt Þxð1Þ ð1Þeaðt1Þ , and in comparison with Eq. (10) we have 4.1. Metrics for evaluating the models
the general solution of Eq. (10) as:
We consider the common evaluation criterion as the fitness
Z t
ð1Þ ð1 Þ aðt1Þ aðt‘Þ ðxÞ function for nonlinear optimization problems, which is also
x ðtÞ ¼ x ð1Þe þ e f ð‘Þd‘; ð20Þ
1
regarded as the key metric to evaluate models. As our previous
knowledge, there is always a certain deviation between the actual
ðxÞ P½‘
where f ð‘Þ ¼ b ‘¼1 tanhðx‘Þ þ c, the trapezoidal rule is employed value and the value predicted by models, the size of this deviation
to estimate the right-hand side of the equation Eq. (20). Noticing is defined as an absolute error and greatly affected by original data.
the boundary value condition xð1Þ ð0Þ ¼ xð0Þ ð0Þ, we have The relative error is considered instead of the absolute error in this
paper, and take the absolute value of relative error to avoid mutual
1 elimination of errors in data sequences, thereby obtaining the
^xð1Þ ðtÞ ¼ xð0Þ ð1Þeaðt1Þ þ absolute percentage error (APE). The APE is defined as:
2
t1 
X 
ðxÞ ðxÞ
 eaðt‘Þ f ð‘Þ þ eaðt‘1Þ f ð‘ þ 1Þ : ð21Þ
 ð0 Þ 
‘¼1 x ðkÞ  ^xð0Þ ðkÞ
APEðkÞ ¼    100% :
 ð22Þ
Then the inverse
accumulated generation operation is xð0Þ ðkÞ


employed to restore X ð1Þ ¼ xð1Þ ð1Þ; xð1Þ ð2Þ;    ; xð1Þ ðnÞ as the predic-
tion results of the HTGM (1,1) model, which be defined in Defini- It’s not difficult to aware APE is only suitable for expressing the
tion 2.3. error for a single data. Furthermore, the calculation rule about
mean absolute percentage error be defined as follows:

4. Optimization of nonlinear parameters by Grasshopper Pn

MAPE ¼ nmþ1
1
k¼m APEðkÞ
Optimization Algorithm  ð0Þ 
Pn x ðkÞ^xð0Þ ðkÞ ; ð23Þ
¼ 1
nmþ1 k¼m  xð0Þ ðkÞ
  100%
From the previous section, we described the modelling mecha-
nism about the HTGM (1,1) and adopted the least square rule to
determine the linear parameters a, bas well as c. Meanwhile, we which describes an entire error when the length of the data
have also noted that the nonlinear parameter x plays a key role sequence is n  m þ 1. Apparently, MAPE would represent different
in developing the HTGM (1,1) model. Therefore, we prepare the meanings when n and m changes. Such as:
nonlinear optimization problem and through the intelligent If m ¼ 1 and n represents the number of modelling data, MAPE
heuristic algorithm to find the optimal x. would evolved to the modelling mean absolute percentage error
In this section, we first developed a unified metric for evaluat- and be denoted as MMAPE.
ing models, whose purpose is to prepare for formulating the non- If m  1 represents the number of modelling data and n repre-
linear optimization problems and validation. The more concept sents the total number of modelling and test data, MAPE would
about these metrics and their calculation rules are projected on evolved to the prediction mean absolute percentage error, and be
Section 4.1. Then we describe the formulation of the nonlinear denoted as PMAPE.
optimization problem and the Grasshopper Optimization If m ¼ 1 and n represents the total number of modelling and
Algorithm applied. The procedure of modelling and calculation is test data, MAPE would evolved to the total mean absolute percent-
summarized in the last subsection. age error, and be denoted as TMAPE.

4
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

4.2. Preparing nonlinear constrained optimization problem Ai ¼ ue^w: ð27Þ

The social ability of the i  th grasshopper is represented as Si
and calculate it as follows:
P n  ð0Þ 
x ðkÞ^xð0Þ ðkÞ
min MMAPE ¼ 1n  xð0Þ ðkÞ   100% X
N


x
8 2 P
k¼1
 3
Si ¼ s xj  xi ; ð28Þ
>
i¼1 tanhðxsÞ þ tanhðxÞ
2
>
> zð1Þ ð2Þ 1
1 j¼1;j–i
>
> 6 2
P  7
>
> 6 ð1Þ P2 7
> where g and u is a constant, ebg and ec
i¼1 tanhðxsÞ þ i¼1 tanhðxsÞ
>
> 6 z ð3Þ 1 3
17 w represent unit vectors in dif-
>
> 6 2 7  
>
>
>
B¼6
6
7
7 ferent directions. N represents the number of grasshopper, xj  xi 
>
> 6 .
.. .
.. .
.. 7
>
> 4 P  5 is the distance between the j  th grasshopper and the i  th
>
> P
>
> i¼1 tanhðxsÞ þ i¼1 tanhðxsÞ
n n1
> zð1Þ ðnÞ 12 1 Grasshopper. The s in Eq. (28) is defined as the social interaction
>
>
>
> 2 ð0Þ 3 function:
>
> x ð2Þ
>
>
>
> 6 xð0Þ ð3Þ 7 r
>
> 6 7 : sðr Þ ¼ fe l  er ; ð29Þ
< Y¼6 7
st: 6 .. 7
> 4 . 5 where f and l represent the intensity of attraction and the endur-
>
>
>
> xð0Þ ðnÞ ance of attraction respectively. Further refinement leads to the final
>
>
>
>  1   form of the mathematical model about the Grasshopper Optimiza-
>
> T
>
> ½a; b; c ¼ BT B BT Y
>
> tion Algorithm shown as follows:
>
>
>
> P½‘ !
>
> ðxÞ
f ð‘Þ ¼ b tanhðx‘Þ þ c X
N
ubd  lbd
 ^
>
>
>
> ‘¼1 X di ¼ c c s x j  xi þ T d ; ð30Þ
>
> 2
>
>
> P  aðt‘Þ ðxÞ
t1
ðxÞ
 j¼1;j–i
>
> ^x ðt Þ ¼ x ð1Þe
ð1Þ ð0Þ aðt1Þ
þ 12 e f ð‘Þ þ eaðt‘1Þ f ð‘ þ 1Þ
>
>
>
> ‘¼1 in which ubd and lbd represent the upper bound and lower bound
:
^xð0Þ ðkÞ ¼ ^xð1Þ ðkÞ  ^xð1Þ ðk  1Þ cd is so far the best solution, c is a shift coefficient
respectively, and T
ð24Þ and be defined as:
It should be noticed that the prediction precision of the model is cmax  cmin
directly affected by the nonlinear parameter x, and it’s necessary c ¼ cmax  l : ð31Þ
L
to improve the performance of the model during modelling by
optimizing the value of x. Therefore, Nonlinear constrained opti-
mization problem is formulated in this paper, which takes 4.4. Modelling and computational procedure
MMAPE as the fitness function and x as the independent variable.
According to the derivation process in section 3, the nonlinearly The modelling and computational procedure about the HTGM
constrained optimization problem is organized as follows. (1,1) model is summarized into following five steps as shown in
Fig. 2.
4.3. The Grasshopper Optimization Algorithm
5. Validation
It would be a challenge to solve the optimization problem with
complicated nonlinear constraints by our conventional mathemat- In this section, we render a series of empirical studies to vali-
ical methods and optimization theory. However, the swarm intel- date the reliability of the HTGM (1,1) model our proposed, and
ligence optimization algorithm is in a better position to solve compared with six classic grey prediction models include the GM
such a tedious problem, which has become quite popular over (1,1) [25], DGM (1,1) [48], NGM (1,1) [49], ARGM (1,1) [32], SAIGM
the past two decades because of the development of computer (1,1) [27] and FGM (1,1) [35], the Grasshopper Optimization Algo-
technology. For the foregoing reasons, the Grasshopper Optimiza- rithm is also employed to optimize nonlinear parameters of the
tion Algorithm (GOA) is adopted to overcome it. FGM (1,1) model. Six practical numerical cases from different liter-
The Grasshopper Optimization Algorithm was proposed by ature are presented as verification in the following content, whose
Shahrzad Saremi in 2017 [42]. As an emerging intelligent optimiza- background is mainly around energy and economy in China. All
tion algorithm, it has drawn full attention once proposed. And it modelling and prediction results are tabulated in Table 1, and more
has been applied to various subject areas for its high efficiency details on them are presented in turn.
and simple operation [43–47]. The Grasshopper Optimization Numerical case 1: (The gas consumption in China) We consider
Algorithm is a nature inspired algorithm, which proposed accord- an example that is mentioned in the literature [41]. The raw data
ing to the social interaction of Grasshopper. from 2008 to 2015 are divided into two groups, the first group
The swarming behaviour of grasshopper can be represented by from 2008 to 2012 is applied to construct the grey model, and
the mathematical model shown as: the remaining group is treated as a validation target. And the com-
parison of the raw data with the results of other models’ is tabu-
X i ¼ r 1 Gi þ r 2 Ai þ r3 Si; ð25Þ lated in Table 1, the results show the MMAPE and PMAPE of the
HTGM (1,1) are much lower than other grey models.
where Xi represents the i  th grasshopper’s current position, Numerical case 2: (The electric load in China) We take the data
r i ði ¼ 1; 2; 3Þ are random number in ½0; 1 in order to provide the from literature [26,28,29] as an example, the data from 2000 to
randomness of behaviour. 2005 as modelling data, and the remaining four data are applied
The gravity the i  th grasshopper born is represented as Gi and to examine the performance of these prediction models. From
it can be calculated as: the modelling and prediction results, we can draw a conclusion
HTGM (1,1) has a more accurate prediction effect in general, even
Gi ¼ g e^g : ð26Þ
in some complex and changeable situations.
The wind advection applied to the i  th grasshopper is Ai and Numerical case 3: (The geothermal, biomass and other con-
calculate it as follows: sumption in China) We regard the data from literature [37] as an
5
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 2. The modelling and computational procedure about the proposed model.

Table 1
Modelling and predicting results of numerical cases in validation.

GM DGM NGM ARGM FGM SAIGM HTGM

Numerical case 1 MMAPE (%) 1.31 2.41 1.43 12.64 0.98 0.96 0.85
PMAPE (%) 12.28 6.20 12.68 16.28 4.32 2.89 2.74
Numerical case 2 MMAPE (%) 0.54 1.06 0.56 11.25 0.36 0.51 0.30
PMAPE (%) 5.97 10.69 6.01 24.06 5.68 7.72 1.74
Numerical case 3 MMAPE (%) 1.78 1.93 1.70 12.12 1.29 0.98 1.22
PMAPE (%) 11.64 4.95 11.49 32.46 4.75 22.29 2.92
Numerical case 4 MMAPE (%) 3.28 3.46 3.27 10.02 3.00 3.28 2.72
PMAPE (%) 18.21 6.96 18.26 7.72 24.23 18.40 3.93
Numerical case 5 MMAPE (%) 2.47 6.17 2.62 12.99 2.10 2.69 1.48
PMAPE (%) 21.88 23.27 22.47 13.15 26.86 19.35 2.44
Numerical case 6 MMAPE (%) 3.31 3.03 3.44 11.44 2.16 2.28 2.15
PMAPE (%) 38.72 22.38 39.38 24.77 14.80 8.75 4.68

example, and construct grey prediction model based on the raw next seven years. All the results about modelling and prediction are
data from 2009 to 2013, then the data from 2014 to 2016 are drawn in Table 1, we can observe that HTGM (1,1) always shows
employed to examine the performance of prediction models. more excellent ability, whether it was modelling or prediction.
Although we noticed the modelling effect of the HTGM (1,1) is Numerical case 6: (The tax revenue in China) We regard the
not the best, and it is not the worst either in comparison with data in the literature [30] as our last examples. The data from
others from Table 1, it much outperforms others in prediction. 2004 to 2010 are employed to construct grey prediction models,
Numerical case 4: (The hydroelectricity consumption in China) then make prediction about behavioural changes over the next
We take the data in the literature [37] as an example, the data from eight years. And we also noticed that although the actual data from
2009 to 2015 are applied to modelling and others from 2016 to 2011 to 2018 show some abnormal trends, what important is this
2018 are applied to validate. Table 1 demonstrate modelling situation is more challenging for other models than HTGM (1,1).
results and prediction effect, it can be noticed that HTGM (1,1)
model showed more excellent performance, whether it was mod-
elling or prediction. 6. Application
Numerical case 5: (The GDP in China) The data from 2000 to
2017 in literature [31] are employed as an example to validate The discharge and treatment of sewage is always a hot topic.
the predictive effect of the HTGM (1,1), we construct different grey Since the party’s 18th National Congress, the ecological civilization
model based on the first 11 data and predict the behaviour over the construction has been regarded as an important content to pro-
6
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

mote our strategic layout and development, and a series of funda- growth in domestic sewage discharge has started to slow in recent
mental and groundbreaking work has been carried out. As a whole, years. Furthermore, since the government gradually attaches
the situation about ecological quality, the discharge and treatment importance to environmental governance, the country has
of sewage in China has continued to improve, showing a trend of increased investment in infrastructure and environmental protec-
steady and positive. But in reality, these achievements are not tion. China’s industrial wastewater treatment industry has been
stable and it would even fall backward once relaxed. rapid development under the support of various policies. Accord-
At present, the ecological civilization construction in China is in ing to our previous statistics, the discharge of industrial wastewa-
a critical period of superimposed pressure and moving forward ter in 2010 is 23.75 billion tons, and it drops to 19.95 billion tons
with heavy burdens, and has entered a critical period of providing by 2015. It can be seen great achievements have been made in
more high-quality ecological products to meet the people’s grow- wastewater treatment in China. From the latest statistics at the
ing needs for a beautiful ecological environment, and has also end of 2017, the industrial wastewater discharge in 2017 reached
reached a window period of having the conditions and ability to 18.16 billion tons, which accounted for 24.13%. It’s quite likely to
solve outstanding ecological environment problems. Therefore, continue the effectiveness of the 12th five-year plan during the
Scientifically and accurately realize the prediction about the sew- 13th five-year plan.
age discharge in China would be conductive to hacking the chal-
lenge about sewage treatment, and it would accelerate the
ecological civilization in China.
6.2. Case1: Forecasting the domestic sewage discharge in China
In this section, HTGM (1,1) model our proposed is employed to
forecast the behaviors about sewage discharge in China, which
In this application case, we mainly focus on the domestic sew-
includes domestic sewage and industrial wastewater. In subsection
age discharge in China from 2006 to 2017, whose development sit-
1, researchers first carried on the preliminary analysis of the sew-
uation can be seen from Fig. 3. The figures from 2006 to 2015 are
age discharge present situation in China. Details about modelling
employed to construct prediction models, where are carried out
and prediction are demonstrated in the middle, and the results of
based on the previous modelling procedure, and the remaining fig-
the demonstration are further discussed at the end.
ures are applied to inspect the predictive performance of models.
The linear parameters can be solved by least square rule, and the
nonlinear parameter x of HTGM (1,1) model is solved by the
6.1. Preliminary analysis about sewage discharge situation in China
Grasshopper Optimization Algorithm. The specific iteration steps
are drawn in Fig. 4 with this condition, where the maximum num-
For a long time, water resources in China have faced a shortage
ber of generations is 100, the number of search agents is 100 and
of resources and serious pollution. What’s more, the discharge of
let x belong to [-1,1]. From Fig. 4, we can observe the fitness func-
sewage is stubbornly high. The raw data about the discharge of
tion of the nonlinear optimization problem converges to the opti-
industrial wastewater and domestic sewage from 2006 to 2017
mal value with only no more than ten iterations when the
are collected from the China Environmental Statistics Year, whose
nonlinear parameter converges to 0.3194.
behavior trends are intuitively drawn in Fig. 3.
The modelling and prediction renderings of all grey prediction
As we can observe, the sewage discharge represented by two
models are drawn in Fig. 5, in which indicates HTGM (1,1) can
different colors shows a diametrically opposite trend. China’s total
more accurately realize the modelling and prediction of domestic
sewage discharge is showing a continuous upward trend. The situ-
sewage discharge. And the metrics listed in previous are employed
ation is probably caused by industrial development and the
to evaluate these models, the evaluation results show modelling
increase of water consumption in our daily life, and these results
mean absolute percentage error (MMAPE) and the prediction mean
against demonstrating the increasingly serious sewage discharge
absolute percentage error (PMAPE) about HTGM (1,1) are 0.7474%,
situation in China.
0.6297% respectively. While those about GM (1,1) are 1.7998% and
As the acceleration of urbanization progress, the discharge of
8.5448%, DGM (1,1) are 2.2905% and 3.5245%, NGM (1,1) are
domestic sewage is increasing and it has gradually become the
1.8182% and 8.5999%, ARGM (1,1) are 9.0612% and 10.3448%,
main source of sewage discharge in China. Happily, the rate of
FGM (1,1) are 1.1916% and 1.7687% as well as SAIGM (1,1) are
1.4285% and 3.9516%.

Fig. 3. The discharge of domestic sewage and industrial wastewater (100 million
tons) in China from 2006 to 2017. Fig. 4. Convergence curves for MMAPE and x by GOA in Application case 1.

7
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

Fig. 5. Modelling and predictive rendering of various grey models in Application

case 1. Fig. 7. Modelling and predictive rendering of various grey models in Application
case 2.
6.3. Case2: Forecasting the industrial wastewater discharge in China
els, which can better describe the behavior of the domestic sewage
Similarly, the same methods and skills are applied in this applica- discharge and the industrial wastewater discharge in China.
tion case to research the industrial wastewater discharge in China, and Meanwhile, the domestic sewage discharge and the industrial
the raw data can be seen in Fig. 3. The prediction models still are con- wastewater in China have been developing in two different direc-
structed by the data from 2006 to 2015, and the left figures are tions. And it is worth mentioning that both their development has
employed to examine their predictive ability and set the same param- been subject to some external interference in recent years, which is
eters condition as Case1. Fig. 6 illustrates the solution process about undoubtedly a challenge for predictive models. In contrast, it is over-
nonlinear parameter x, and it shows x converges to 0.1011 after come smoothly by HTGM (1,1) model proposed. Above all, we have
less than ten iterations with the Grasshopper Optimization Algorithm. reasons to believe that the HTGM (1,1) model is a grey model more
Using the metrics for evaluating the models we can noticed that suitable for predicting the discharge of the sewage industry in China.
HTGM (1,1) model’s modelling mean absolute percentage error
(MMAPE) and its prediction mean absolute percentage error (PMAPE) 7. Conclusion
are 0.8914% and 1.2158%, while those about other existing models
are 1.3802% and 5.5312%, 3.9153% and 2.5645%, 1.3816% and In this work, a novel hyperbolic time-delayed grey model called
5.5225%, 9.7716% and 24.4688, 0.9454% and 1.8440%, 0.9002% and the HTGM (1,1) model was presented. During the stage of modelling,
1.7884%. Moreover, all the results are visually drawn in Fig. 7 and the popular Grasshopper Optimization Algorithm was employed to
show HTGM (1,1) model is obviously better than others. determine the nonlinear parameter of the HTGM (1,1) model. A series
of practical numerical cases were presented to verify the reliability of
HTGM (1,1) model in comparison with six classical grey prediction
6.4. Further discussion
models including GM (1,1), DGM (1,1), NGM (1,1), ARGM (1,1),
FGM (1,1) and SAIGM (1,1). Then we applied HTGM (1,1) to predict
As can be seen from the previous application cases, our pro-
the sewage discharge in China, and those results again demonstrated
posed HTGM (1,1) model always provides a satisfactory modelling
HTGM (1,1) model outperformed other grey models.
capability compared to the other six classical grey prediction mod-
At present, the domestic sewage discharge and the industrial
wastewater discharge in China are growing in two directions,
and the government’s policies on controlling industrial wastewater
discharge have reached remarkable achievements. At the same
time, the potential threat posed by domestic sewage should not
be ignored. The occurrence of environmental pollution is closely
related to people’s production and daily life. Although it is impos-
sible to completely eliminate it, with the increase of people’s atten-
tion and the improvement of technology, the pollution will be
controlled to a certain extent. The construction of ecological civi-
lization will remain the theme of long-term development in the
future. All in all, the treatment and discharge of sewage is a
long-term behavior, and will continue throughout the entire pro-
cess of economic development in the future.

Declaration of Competing Interest

The authors declare that they have no known competing finan-

cial interests or personal relationships that could have appeared
Fig. 6. Convergence curves for MMAPE and x by GOA in Application case 2. to influence the work reported in this paper.
8
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

Acknowledgments [25] Liu S, Forrest JYL. Grey systems: theory and applications. Springer Science &
Business Media; 2010.
[26] Xu N, Dang Y, Gong Y. Novel grey prediction model with nonlinear optimized
We thank anonymous peer reviewers for carefully revising our time response method for forecasting of electricity consumption in China.
manuscript and for his or her useful comments. In addition, our Energy 2017;118:473–80.
[27] Zeng B, Meng W, Tong M. A self-adaptive intelligence grey predictive model
research work was supported by the National Natural Science
with alterable structure and its application. Eng Appl Artif Intell
Foundation of China (No. 71901184, 71771033, 71571157), 2016;50:236–44.
Humanities and Social Science Fund of Ministry of Education of [28] Ding S, Hipel KW, Dang Y. Forecasting China’s electricity consumption using a
new grey prediction model. Energy 2018;149:314–28.
China (No. 19YJCZH119), the Open Fund (PLN201710) of State
[29] Zhao H, Guo S. An optimized grey model for annual power load forecasting.
Key Laboratory of Oil and Gas Reservoir Geology and Exploitation Energy 2016;107:272–86.
(Southwest Petroleum University), and National Statistical Scien- [30] Yu L, Xiang X, Yang L. A new background value improvement of fractional
tific Research Project (2018LY42). order accumulated FAGM (1, 1) model and its application. Asian Res J Math
2019:1–14.
[31] Yang Y, Wei HE. The predietion and analysis of GDP in China based on
combination forecast model. Sci Technol Ind 2011;3.
References
[32] Wu L, Liu S, Chen H, et al. Using a novel grey system model to forecast natural
gas consumption in China. Math Problems Eng 2015;2015.
[1] Ma M, Ma X, Cai W, et al. Low carbon roadmap of residential building sector in [33] Zeng B, Li C. Improved multi-variable grey forecasting model with a dynamic
China: Historical mitigation and prospective peak. Applied Energy 2020;273 background-value coefficient and its application. Comput Ind Eng
(115247). 2018;118:278–90.
[2] Chen L, Cai W, Ma M. Decoupling or delusion? Mapping carbon emission per [34] Ma X. A brief introduction to the Grey Machine Learning. Journal of Grey
capita based on the human development index in Southwest China. Science of System 2019;31(1):1–12.
The Total Environment 2020;138722. https://doi.org/10.1016/j.scitotenv. [35] Wu L, Liu S, Yao L, et al. Grey system model with the fractional order
2020.138722. accumulation. Commun Nonlinear Sci Numer Simul 2013;18(7):1775–85.
[3] Singh P, Kansal A, Carliell-Marquet C. Energy and carbon footprints of sewage [36] Ma X, Xie M, Wu W, et al. The novel fractional discrete multivariate grey
treatment methods. J Environ Manage 2016;165:22–30. system model and its applications. Appl Math Model 2019;70:402–24.
[4] Suchetana B, Rajagopalan B, Silverstein JA. Investigating regime shifts and the [37] Wu W, Ma X, Zeng B, et al. Forecasting short-term renewable energy
factors controlling Total Inorganic Nitrogen concentrations in treated consumption of China using a novel fractional nonlinear grey Bernoulli
wastewater using non-homogeneous Hidden Markov and multinomial model. Renewable Energy 2019;140:70–87.
logistic regression models. Sci Total Environ 2019;646:625–33. [38] Ma Xin, Wenqing Wu, Zeng Bo, Wang Yong, Xinxing Wu. The conformable
[5] Suchetana B, Rajagopalan B, Silverstein JA. Assessment of wastewater fractional grey system model. ISA Trans 2019;2019(07):009.
treatment facility compliance with decreasing ammonia discharge limits [39] Hu Y et al. Forecasting manufacturing industrial natural gas consumption of
using a regression tree model. Sci Total Environ 2017;598:249–57. China using a novel time-delayed fractional grey model withmultiple
[6] Ding L, Lv Z, Han M, et al. Forecasting China’s wastewater discharge using fractional order. Computational and Applied Mathematics 2020;39(4):1–30.
dynamic factors and mixed-frequency data. Environ Pollut 2019;255:113148. https://doi.org/10.1007/s40314-020-01315-3.
[7] Alkhudhiri A, Darwish NB, Hilal N. Analytical and forecasting study for [40] Ma X, Mei X, Wu W, et al. A novel fractional time delayed grey model with
wastewater treatment and water resources in Saudi Arabia. J Water Process Grey Wolf Optimizer and its applications in forecasting the natural gas and
Eng 2019;32:100915. coal consumption in Chongqing China. Energy 2019;178:487–507.
[8] Ying L, Li-Kun Y, Jing Z. Study the ways to forecast the discharge of restaurant [41] Ma X, Liu Z. Application of a novel time-delayed polynomial grey model to
wastewater in Beijing. Procedia Environ Sci 2011;11:850–7. predict the natural gas consumption in China. J Comput Appl Math
[9] Man Y, Hu Y, Ren J. Forecasting COD load in municipal sewage based on ARMA 2017;324:17–24.
and VAR algorithms. Resour Conserv Recycl 2019;144:56–64. [42] Saremi S, Mirjalili S, Lewis A. Grasshopper optimisation algorithm: theory and
[10] Liu Y, Ganigué R, Sharma K, et al. Event-driven model predictive control of application. Adv Eng Softw 2017;105:30–47.
sewage pumping stations for sulfide mitigation in sewer networks. Water Res [43] Luo J, Chen H, Xu Y, et al. An improved grasshopper optimization algorithm
2016;98:376–83. with application to financial stress prediction. Appl Math Model
[11] Li J, Sharma K, Liu Y, et al. Real-time prediction of rain-impacted sewage flow 2018;64:654–68.
for on-line control of chemical dosing in sewers. Water Res 2019;149:311–21. [44] Xiang X, Ma X, Wu W, et al. Research and application of novel Euler
[12] Wei X, Kusiak A. Short-term prediction of influent flow in wastewater polynomial-driven grey model for short-term PM10 forecasting. Grey
treatment plant. Stoch Environ Res Risk Assess 2015;29(1):241–9. Systems: Theory and Application 2020. https://doi.org/10.1108/GS-02-2020-
[13] Ansari M, Othman F, El-Shafie A. Optimized fuzzy inference system to enhance 0023.
prediction accuracy for influent characteristics of a sewage treatment plant. [45] Łukasik S, Kowalski PA, Charytanowicz M, et al. Data clustering with
Sci Total Environ 2020:137878. grasshopper optimization algorithm. In: 2017 Federated conference on
[14] Zhou Y. Real-time probabilistic forecasting of river water quality under data computer science and information systems (FedCSIS). IEEE; 2017. p. 71–4.
missing situation: Deep learning plus post-processing techniques. J Hydrol [46] Fathy A. Recent meta-heuristic grasshopper optimization algorithm for
2020:125164. optimal reconfiguration of partially shaded PV array. Sol Energy
[15] Zhao G, Li N, Li B, et al. ANN model for predicting acrylonitrile wastewater 2018;171:638–51.
degradation in supercritical water oxidation. Sci Total Environ [47] Ahanch M, Asasi MS, Amiri MS, Grasshopper A. Optimization Algorithm to
2020;704:135336. solve optimal distribution system reconfiguration and distributed generation
[16] Ma M, Ma X, Cai W, et al. Carbon-dioxide mitigation in the residential building placement problem. In: 2017 IEEE 4th international conference on knowledge-
sector: A household scale-based assessment. Energy Conversion and based engineering and innovation (KBEI). IEEE; 2017. p. 0659–66.
Management 2019;198(11915). [48] Xie N, Wang R, Chen N. Measurement of shock effect following change of one-
[17] Guo H, Jeong K, Lim J, et al. Prediction of effluent concentration in a child policy based on grey forecasting approach. Kybernetes 2018;47
wastewater treatment plant using machine learning models. J Environ Sci (3):559–86.
2015;32:90–101. [49] Cui J, Liu S, Zeng B, et al. A novel grey forecasting model and its optimization.
[18] Du P, Wang J, Yang W, et al. A novel hybrid model for short-term wind power Appl Math Model 2013;37(6):4399–406.
forecasting. Appl Soft Comput 2019;80:93–106.
[19] Du P, Wang J, Yang W, et al. Container throughput forecasting using a novel
hybrid learning method with error correction strategy. Knowl-Based Syst
2019;182:104853. Xiwang Xiang was born in 1998 at Sichuan of China. He
[20] Du P, Wang J, Guo Z, et al. Research and application of a novel hybrid is currectly studying Southwest University of Science
forecasting system based on multi-objective optimization for wind speed and Technology, and was awarded the National
forecasting. Energy Convers Manage 2017;150:90–107. Encouragement Scholarship in China in 2018.
[21] Yang W, Wang J, Lu H, et al. Hybrid wind energy forecasting and analysis
system based on divide and conquer scheme: A case study in China. J Cleaner
Prod 2019;222:942–59.
[22] Wang Y, Zhang C, Chen T, et al. Modelling the nonlinear flow for a multiple-
fractured horizontal well with multiple finite-conductivity fractures in triple
media carbonate reservoir. J Porous Media 2018;21(12).
[23] Wang Y, Tian D, Li G, et al. Dynamic analysis of a fractured vertical well in a
triple media carbonate reservoir. Chem Technol Fuels Oils 2019:1–10.
[24] Wang Y, Yi X. Flow modelling of well test analysis for a multiple-fractured
horizontal well in triple media carbonate reservoir. Int J Nonlinear Sci Numer
Simulat 2018;19(5):439–57.

9
X. Xiang et al. Ain Shams Engineering Journal xxx (xxxx) xxx

Xin Ma was born in 1989 at Sichuan of China. He Wenqing Wu was born in 1986 at Sichuan of China. He
received his BS degree in Mathematics and Applied received his BS degree in Mathematics and Applied
Mathematics from Xihua University in 2010, MS and Mathematics from Southwest University of Science and
PhD degree in Petroleum Engineering (Computing Technology, MS and PhD degree in Applied Mathemat-
Technology) from Southwest Petroleum University. He ics from Sichuan Normal University. He started to work
was a visiting scholar of KU Leuven during Aprial and as a lecturer in Southwest University of Science and
July in 2019. He is currectly working as an associate Technology from 2015. His research interest includes
professor in Southwest University of Science and Tech- Stochastical Operations, grey system theory and appli-
nology. His research interest includes Grey Systems and cations.
Machine learning.

Yizhu Fang was born in 1989 at Sichuan of China. She Gaoxun Zhang was born in December 1977. He
received her BS degree in Petroleum Engineering from received the BS degree from Henan Normal University
Southwest Petroleum University, MS degree in Oil-Gas (2004), the MS degree from Chengdu University of
Field Development Engineering from Southwest Petro- Information Technology (2007), and he received his Ph.
leum University. She started to work as a engineer in D. from the University of Electronic Science and Tech-
Exploration and Development Research Institute, Pet- nology of China. His research interests are in the areas of
rochina Southwest Oil & Gasfield Company from 2015. financial engineering and investment research, espe-
Her main research interests focus on medium and long cially theory and application of option pricing.
term development planning of oil and gas fields.

10