Application of coal price prediction method based

on ISSA-LSSVR method in state optimization design

of inclined seam open-pit mine
Bo Cao 
Liaoning Technical University
Shuai Wang  (  [email protected] )
Liaoning Technical University
Bo Zhao 
Liaoning Technical University
Qingyi Li 
Liaoning Technical University
Mingjia Lv 
Liaoning Technical University
Guangwei Liu 
Liaoning Technical University

Article

Keywords: open-pit mine, coal price volatility, Improved sparrow search algorithm (ISSA), Least square
Support Vactor Regression (LSSVR), state optimization

Posted Date: December 15th, 2022

DOI: https://doi.org/10.21203/rs.3.rs-2339783/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.  
Read Full License

Additional Declarations: No competing interests reported.

Application of coal price prediction method based on
ISSA-LSSVR method in state optimization design of
inclined seam open-pit mine
Bo Cao, Shuai Wang*, Bo Zhao, Qingyi Li, Mingjia Lv ,Guangwei Liu
College of mining, Liaoning Technical University, Fuxin 123000, China
*
Correspondence: [email protected]; Tel.: +86-13833435405

Abstract: As an important link in the complex system engineering project of open-pit mining, the quality of boundary
determine the performance of the project to a large extent. However, the traditional design method cannot effectively
measure the impact of uncertainties on the realm optimisation process. In this article, a coal price time series forecasting
model that considers the amount of redundancy is proposed, which combines an improved sparrow search algorithm
(ISSA) and a least squares support vector regression machine regression algorithm (LSSVR). The optimal values of the
penalty factor and kernel function parameter of the LSSVR model are selected by ISSA, which improves the prediction
accuracy and generalisation performance of the forecasting model. A multi-step decision optimisation method under
fluctuating coal price conditions is proposed, and the model prediction results are applied to the boundary optimisation
design process. Using the widely applied block model as the basis, a set of optimal production nested pits is obtained
and only obtained, allowing the realm design results to fit the coal price fluctuation trend and further enhance the
enterprise efficiency. The applicability and effectiveness of the proposed method is verified using an ideal
two-dimensional inclined coal seamopencast mine model as an example.
Keywords: open-pit mine, coal price volatility, Improved sparrow search algorithm (ISSA), Least square Support
Vactor Regression (LSSVR), state optimization.

The open-pit mining method involves drilling layers of earth to expose ore and minerals while
iteratively deepening the pit [1]. Excavated material from the ore and waste rock pits is loaded onto trucks
and transported to specific designated areas [2]. The production and operating space of open pit mine is
constantly changing over time. Besides, the production process is complex and involves numerous
segments, including boundary definition, engineering organisation and schedule design. The size (or
extent) as well as the sequence (or scheduling) of extraction are the most critical decisions within the
domain of strategic or long-term production planning for open pit mining operations [3]. Once the final
boundary has been defined, the design of the deposit and the mining schedule are based on the mine's
economic and technical capabilities.
The traditional open-pit mine boundary optimization design method usually uses block model to
represent the mine as a series of blocks with predetermined economic value, and delineates the mining
scope with the goal of maximum net present value. An economic block model that constitutes thousands
of ore and waste blocks becomes the basic input to the strategic mine planning process that defines the: (i)
mining method (surface vs. underground); (ii) the size or extent of mining, and; (iii) the sequence of
extraction [4]. Typically, mining ranges consist of millions of blocks and face multiple selection decisions.
The conventional method only applies the known and constant data; however, the uncertainty from
commodity prices and geological aspects are ignored in the simulation [5].
The conventional approach achieves the phase design and ultimate pit limit in two separate steps.
The first step is to result in a series of nested pits. The second step utilizes these nested pit shells to
develop a number of phases within the ultimate pit [6] The basic idea of nested pits is that given a series of
non-negative income factors β1<β2<...<βk, to solve the optimization problem of open pit under this
condition. The solution of this problem meets the nesting property, that is, the pit obtained by using
income factor βi is completely included in the pit obtained by using income factor βi+1. However, the
effectiveness of this approach has been questioned: (1) when the economy is in a good state, some
benefits may be lost due to less mining of coal resources; (2) When the economic situation is poor, some
of the resources within the original limit cannot bring the expected benefits, and the mining of this part of
coal will lead to the reduction of economic benefits.
Among many economic indicators, coal price plays a decisive role in the optimization of open pit
mining limits. China is the largest coal producer and coal user in the world. The change of coal price is
restricted by many factors including international and domestic dimensions, supply and demand
relationship and macro policy. As a commodity, the price of coal is greatly affected by the external
environment, which is directly reflected in its strong volatility. Hence, scientific and accurate coal price
prediction assists relevant enterprises to anticipate the market trend and to prudently adjust economic
activities [7].
Coal price forecasting can be categorized as an onerous time series modeling task. Different from
ordinary regression, coal price forecasting needs to utilize the multiple input characteristics of the coal
price in a past period to predict its future trends [8]. Coal price forecasting methods in the literature can be
roughly divided into two categories, namely statistical models and artificial intelligence (AI) models. The
former is based on conventional econometrics. Representative models primarily include vector
[9] [10]
autoregression (VAR) , autoregressive integrated moving average (ARIMA) , and generalized
[11]
autoregressive conditional heteroskedasticity (GARCH) . In recent years, thanks to the improvement of
related algorithms and computing techniques, AI models have achieved major technological
breakthroughs in energy price forecasting. The early recommended AI models are machine learning
algorithms such as support vector regression (SVR) [12], random forest (RF) [13], and artificial neural
networks (ANN) [14].
As a typical machine learning method, the support vector machine (SVM) method is calibrated
utilizing a structural risk minimization principle to minimize an upper bound of the generalization error
[15]
. Although SVM has been widely applied in hydrological forecasting, some drawbacks like low
[16]
execution efficiency and high computation cost may exist . To overcome these defects, the
least-squares support vector machine (LSSVM) method was proposed. Compared to SVM, LSSVM
adopts linear equality constraints rather than quadratic programming to solve complex problems and can
get better model performance [17].
SVR is an important application branch of support vector machine. The SVR method structure is
more straightforward than fuzzy and ANN models that enhance the predicting model [18]. Although the
SVR application has various advantages, it has some unknown parameters in its structure, which
drastically affect the prediction accuracy and generalization performance. Terefore, optimization of the
model parameters is required to obtain a prediction model with good performance [19]. Many scholars
have used metaheuristic algorithms for model parameter optimization due to their simple structure, fast
efficiency, and few adjustment parameters [20].
Sparrow search algorithm (SSA) is a new metaheuristic algorithm proposed by Xue and Shen in
[21]
2020 . Sparrow search algorithm has been a wide concern because of its high efficiency, high
convergence accuracy, and strong stability. However, SSA, like other population intelligence optimization
algorithms, still suffers from the problems of decreasing population diversity and easily falling into local
optimality when its search is close to the global optimum. In order to overcome the shortcomings of
traditional SSA, many scholars have improved SSA by increasing the initial population diversity and
escaping from the local optimum. Tent chaotic mapping is used to initialize the population so that the
initial individuals are distributed as evenly as possible [22], while introducing Gaussian variation and
chaotic perturbation to help individuals jump out of the local optimum, thus overcoming the drawback
that SSA is prone to fall into the local optimum. Mao et al [23] integrated the idea of sine-cosine algorithm
to balance the local and global searchability in the discoverer position update method of SSA and
introduced Levy flight strategy in the follower position update method to perturb the variation of the
current optimal solution and strengthen the local escape ability, which obviously improved the efficiency
of SSA solution. Tang et al [24] used cubic mapping to initialize the population to obtain an improved
chaotic sparrow search algorithm, while using the Gaussian wandering strategy to help the algorithm
jump out of stagnation.
Considering the traversal uniformity and fast convergence of the Skew-Tent mapping and the better
local searchability of the adaptive search strategy. Based on the improved sparrow search algorithm
(ISSA) proposed by Lv et al [22], a hybrid model of ISSA-based LSSVM (ISSA-LSSVM) was proposed
for coal price forecasting. The optimal values of the penalty factor and kernel function parameter of the
LSSVR model are selected by ISSA, and the model prediction results are applied to the boundary
optimisation design process. Using the widely applied block model as the basis, a set of optimal
production nested pits is obtained and only obtained, allowing the realm design results to fit the coal price
fluctuation trend within the redundancy range and further enhance the enterprise efficiency.
The rest of the article is organized as follows: the basic principle of the LSSVM and ISSA methods
are presented and the hybrid model ISSA-LSSVM is briefly illustrated. The details of the study case and
data and develops forecast models are then introduced, before the statistics of the forecast results are
presented and the results discussed. The applicability and effectiveness of the proposed method is verified
using an ideal two-dimensional inclined coal seam opencast mine model as an example. Finally, the
conclusion is presented.

Improvement of Sparrow search algorithm ISSA

Classic Sparrow Search Algorithm. Sparrow Search Alogorithm (SSA) was proposed by Xue 21 in 2020,
which is mainly based on foraging behavior and anti-preying behavior of sparrow. Sparrows like to live in
groups and have a strong memory. There is a clear division of labor within their species, and some of
them are responsible for finding foraging areas and directions for the species. They are called discoverers.
When predators appear around the population, sparrows that are aware of the danger will give an alarm
signal, and the whole population will immediately take anti-preying behavior to avoid the danger [22].
Using virtual sparrows to find food by idealizing their behavior. The sparrow's position can be
represented by the following matrix.
 X 1   x11 x12 x1d 
  x x x 
X 2   21 2d 
X =
22
= （1）
   
   
 X N   xn1 xn 2 xnd 
Where n is the number of sparrows and d is the dimension of the variable to be optimized. Then the
fitness values of all individuals can be represented by the following vectors.
  f 1   f ( x11 x12 x1d ) 
 f  f x 
 ( 21 x22 x2 d ) 
FX =  2  =   （2）
 
   
 f N   f ( xn1 xn 2 xnd ) 

The producers usually have high levels of energy reserves and provide foraging areas or directions
for all joiners. It is responsible for identifying areas where rich food sources can be found. The level of
energy reserve depends on the assessment of the fitness value of the individual. Producers with higher
fitness values get the food first in the search process. In addition, producers are responsible for finding
food and directing the movement of the entire population. Therefore, producers can search for food in a
wider range of places than joiners. During each iteration, the producer's location is updated by the
formula.

 t  −i 
 xij  exp   if R2  ST
xijt +1 =    Tmax  （3）
 t
 xij + Q  L if R2  ST

where t denotes the current iteration, j=1,2,…,d. xijt denotes the value of the jth dimension of the ith

sparrow at iteration t. Tmax is the constant with the highest number of iterations. α∈(0,1] is a
random number. R2(R2∈[0,1]) and ST(ST∈[0.5,1.0]) denote the alarm value and the safety threshold,
respectively. Q is a random number that obeys a normal distribution. L denotes a matrix of order 1×d.
where each element is 1. When R2<ST, indicating that there are no predators around, the producer
enters the wide-area search mode. If R2≥ST, it means that some sparrows have found predators and
all sparrows need to fly quickly to other safe areas.
Sparrows with higher energy levels will act as producers. A few hungry joiners are more likely to fly
elsewhere for more energy. Participants seek out food by following the producers who provide the best
food. At the same time, some subscribers may constantly monitor producers and compete for food to
increase their own predation rates. The position update formula of the subscriber is
  x tworst − xijt 
Q  exp   if i  n / 2
  i2 
xijt +1 =    （4）
 t +1 t +1 +
 x p + xij − x p  A L
t
otherwise

where xP is the optimal position occupied by the producer. xworst denotes the current global worst
position. A denotes a 1×d-dimensional matrix where each element is randomly assigned 1 or -1 and A+
=AT(AAT)-1. When i>n/2, this indicates that the i-th accession with the worse fitness value is most likely
not to be allocated food.
Assuming that the number of sparrows aware of danger is 10%-20% of the total population, and that
the initial positions of the above sparrows are generated randomly in the population, sparrows at the edge
of the group will move quickly to a safe area to gain a better position when they are aware of danger,
while sparrows in the middle of the group walk randomly to get closer to others, the mathematical model
can therefore be expressed as
  x tbest +   xijt − xbest
t
if fi  f g


xijt +1 =   xijt − xworst
t  （5）
 xij + K    if fi = f g
t

  ( fi − f w ) +  
  

where xbest is the current global optimal position. The random number β~N(0,1) is the step control

parameter. K   −1,1 is the random number. fi is the current fitness value of the sparrow. fg and fw are

the current global best and worst fitness values, respectively. ε is the minimum constant to avoid zero
division error. When fi ≥ fg , the sparrow is at the edge of the population and vulnerable to predators; when
fi = fg, the sparrow is in the middle of the population. Once a sparrow is aware of a predator threat, it will
approach other sparrows and adjust its search strategy to avoid being attacked.

Improved Sparrow Search Algorithm. Although SSA has the advantages of fast convergence speed,
high stability, few adjustment parameters, simple calculation, etc., it also has the disadvantages of falling
into local optimal as other intelligent search algorithms. Therefore, the SSA algorithm is improved in this
article. Considering the better local search ability of Gaussian distribution and the characteristics of
uniform traversal and fast convergence of Tent chaotic sequence, this article selects the chaotic sparrow
search optimization algorithm proposed by Xin Lv et al 22, which uses Tent chaotic mapping to initialize
the population, making the initial individuals as uniformly distributed as possible, and introduces
Gaussian variation and chaotic perturbation when the population appears to "gather" or "diverge". "The
Tent chaos mapping is used to initialize the population so that the initial individuals are as evenly
distributed as possible.

tent chaotic mapping. Chaos, as a non-linear phenomenon prevalent in nature, has been applied to
optimization search problems by many scholars because of the randomness, ergodicity and regularity of
chaotic variables, which can not only effectively maintain the diversity of the population, but also help
the algorithm to obtain the global optimal solution and effectively prevent local convergence. Common
logistic mappings have a high probability of taking values in the ranges [0,0.05] and [0.9,1], so the
algorithm's optimisation search speed is affected by the unevenness of the logistic traversal and the
optimisation search efficiency will be reduced. (1) The study shows that the traversal uniformity and
convergence speed of Tent mapping are better than those of Logistic mapping. Meanwhile, in order to
avoid its falling into small or unstable periodic points without destroying the three characteristics of
chaotic variables, Zhang Na et al. () introduced random variables rand(0,1)×1/NT on the original
expression of Tent mapping, and the improved expression is
 1 1
 2 zi + rand ( 0,1)  N 0 z
2
zi +1 = 
T
（6）
 2 (1 − z ) + rand ( 0,1)  1 1
 z 1
 i
NT 2
The expression after the Bernoulli transformation is
1
zi +1 = ( 2 zi ) modl + rand ( 0,1)  （7）
NT
 where: NT is the number of particles within the chaotic sequence; rand(0,1) is the random number
between [0,1].
According to the properties of the Tent mapping, the steps to generate a chaotic sequence in the
feasible domain are as follows:
Step 1 Randomly generate the initial value z0 within (0,1) , noting i=0.
Step 2 Iterate using equation (7) to produce a sequence of Z with i self-increasing by 1.
Step 3 If the maximum number of iterations is reached, the programrun stops and the resulting Z
sequence is saved.

Tent chaotic perturbations and Gaussian variation. Chaotic perturbation is introduced to avoid it
falling into a local optimum and to improve the global search capability and the accuracy of the search for
an optimum. The steps of chaotic perturbation are described as follows.
Step 1 Apply equation (7) to generate the chaotic variable Zd .
Step 2 Carry the chaotic variables into the solution space of the problem to be solved:
d
X new = d min + ( d max − d min ) Z d （8）

d
Where dmin and dmax are the minimum and maximum values of the d-dimensional variable X new ,

respectively.
Step 3 Chaotic perturbation of the individual according to equation (9):

 = ( X  + X new ) / 2
X new （9）

Where: X  is the individual to be chaotically perturbed; X new is the amount of chaotic

 is the individual after the chaotic perturbation.
perturbation generated; X new
The Gaussian variation is derived from the Gaussian distribution, specifically when the variation
operation is performed by replacing the original parameter value with a random number that fits a normal
distribution with mean µ and variance σ2 . The variance formula is

mutation ( x ) = x (1 + N ( 0,1) ) （10）

Where: x is the original parameter value; N(0,1) denotes a normally distributed randomnumber with
expectation 0 and standard deviation 1; mutation(x) is the value after Gaussian variation. The Gaussian
distribution has a strong local search capability, and for optimization problems with a large number of
local minima, it is beneficial for the algorithmto find global minima with high efficiency and accuracy,
and also improves the robustness of the algorithm.

Implementation. Tent chaos search and Gaussian variation are introduced into the Sparrow
algorithm, which is implemented as Figure 1.
 Sort, and pick the first N Calculate the fitness
Begin individuals value fi

Initialize the Output result

population

N
t≤tmax？ End

Y
Calculate individual fitness
and location
t=t+1
Calculate the location of the
discover and the joiner

Update the position Update the position Update the position

of the discoverer of the joiner of the watcher

Calculate Individual fitness

value fi and Average fitness
value of population favg

Performed tent N Y Perform Gaussian

chaotic fi＜favg？
variation
disturbance

the disturbance N The original N the mutation is

is better than the better than the original
original? remains the same individual?

Y Y
The disturbed individual The mutated individual
replaces the original replaces the original

Update the optimum position,

the worst position and the
fitness value of the
population

Figure 1. Flow chart of the ISSA

Least square vector support machine method

SVR, as a branch derived from SVM, was proposed by Harris [25]. The output of SVR is discrete
values and by introducing the insensitive parameter ε, a new class of models with continuous values for
the output is obtained, namely the Support Vector Machine Regression (SVR) models. For prediction
problems with small samples, the SVR model is used with good results, as shown in Figure 2.

i

 f ( x) + 

f ( x)

f ( x) − 


 *i

Figure 2. SVR 2D schematic

The least squares support vector machine (LSSVM) method is an improvement on the traditional
SVM and has unique advantages in dealing with small samples and non-linear problems. lSSVM reduces
computational complexity to some extent and increases computational speed.
 Given m training samples D = ( x1 , y1 ) , ( x2 , y2 ) , , ( xm , ym ) xi  R n, yi  R , it is desirable to

obtain a linear regression function f ( x ) = w   ( x ) + b (where ω is the weight vector and b is the bias)

such that it is as close as possible to the y-value at each point, with losses calculated only when

f ( x ) − y   . This is equivalent to constructing an interval band of width 2ε with the original function

f(x) as the median, and if the fitted value yi falls into the interval band it is considered to be a correct
prediction and no loss is calculated. It can therefore be considered as providing some redundancy for the
variation in the data during the time prediction process.
LSSVM can be used to map the input space to a high-dimensional feature space in time series
regression with a non-linear function that can be described as

 1 1
m

 2
2
 min w + C i y i
2 
i =1
*

 （11）
 yi    ( xi ) + b  = 1 − i
 s.t.
 i = 1, 2, , m

where C>0 is the penalty factor, i is the error term.

To solve Equation (12), the Lagrange operator is introduced, and the corresponding Lagrange
function is constructed as
m m
1 1
2 2 i =1 
L (, b, i ,  ) =  2 + C i2 −  ( y (   ( x ) + b ) − 1 +  )
i =1
i i i i （12）

where: αi is the Lagrange multiplier.

According to the KKT condition gives the partial differentiation of , b, i ,  in equation (12).

 L
m

 

= 0   − i ( xi ) = 0
i =1
 m
 L
 b
= 0   i = 0
（13）
 i =1
 L
 = 0  Ci − i = 0
 i
 L
 = 0     ( xi ) + b + i − yi = 0
  i
This can be simplified by eliminating as follows by eliminating  and i .

0 eT 
  b   0 
 e  + 1 E  a  =  y  （14）
    
 

Where: eT = 1, , L , y =  y1,

T
( ) T
( )
, y m  , ij = K xi  x j =  ( xi )  x j , i = 1, 2, ,m .

The role of kernel functions K xi  x j ( ) in SVR is to be able to map sample data from a

low-dimensional space to a high-dimensional space without guaranteeing an increase in its inner product
operations, enabling the process of processing non-linear samples to linearisation. Several common types
of kernel functions are: linear kernel functions, polynomial kernel functions, Gaussian radial basis (RBF)
kernel functions, and multilayer perceptron kernel functions.
The selection of kernel function is crucial to the effectiveness of the support vector machine model.
The most widely used function in this paper is the Gaussian radial basis kernel function, whose expression
is shown in Equation (15).

 x − xi2 
( )
K xi  x j = exp  − 
g 2 
（15）


where g>0 is the bandwidth of the Gaussian kernel and the kernel function parameters.
LSSVM modelling requires the setting of two key parameters: the penalty factor C and the kernel
function parameter g. If the value of C is large, LSSVM overlearns; if the value of C is small, LSSVM
underlearns. g is too large, LSSVM model accuracy decreases; g is too small, LSSVM model generalisation
ability decreases. The magnitude of the above two parameters determines the accuracy and the amount of
redundancy of the time series prediction, and plays a crucial role in the prediction effect. In general, the
selection of penalty factor C and kernel function parameter g is still mainly based on human judgement,
which may lead to distinctly different results due to different levels of decision makers. Therefore, this
article proposes to use the ISSA a lgorithm to predict reasonable C and g values for the SVR model, and
then back-calculate the insensitive parameter ε to determine the amount of redundancy in the prediction
period, so as to improve the prediction accuracy and generalisation performance of the prediction model.

4. ISSA-LSSVR coal price time series forecast

Construction of time series prediction model. In this article, we propose to use the LSSVR model for
time series prediction. The performance of the LSSVR prediction model is mainly determined by the
penalty factor C and the kernel function broadband g. Blind selection of parameters easily leads to the
problems of low accuracy and low efficiency of the prediction model. ISSA has the advantages of high
convergence ac curacy, fast optimization and good stability. In this article, an ISSA-LSSVR prediction
model is constructed, and the parameters of the LSSVR prediction model are optimised by using ISSA to
search for the combination of penalty factor C and kernel function parameter g. The flow chart is shown
in Figure 3.
 Begin
Divided the population into
discoverers and Joiners

Data input and normalization

Update the location of finder
and entrant
(SVR modeling location)

Construct the prediction model of

LSSVR
Sensing predators and
updating population positions

Initialize the sparrow population

and set the individual position
as(Ci,gi) Calculate the optimal fitness
and update the optimal
individual position

Output the optimal Y Meet the termination N

parameters C、g criteria?

LSSVM prediction
LSSVR model End
model

Figure 3. ISSA-LSSVR process

Coal price data collection. This article uses the Prospective Database website (https://d.qianzhan.com/)
to collect raw coal prices and power coal prices for the ten-year period 2010 to 2020, the trends of which
are compared in the Figure 4.
1000 1200
Thermal coal price ( yuan / ton )

900 Pithead coal price 1100

Pithead coal price ( yuan / ton )

Thermal coal price 1000

800
900
700
800
600 700
500 600
400 500
400
300
300
200
200
100 100
0 0
2010/1/1 2012/1/1 2014/1/1 2016/1/1 2018/1/1 2020/1/1

Figure 4. Comparison of coal pithead price and thermal coal price trend
As shown in Figure 4, the trends of raw coal pit prices and the value of power coal show a strong
correlation, i.e. the trends are essentially the same for both. For surface mines, where companies obtain
their economic value through the direct sale of raw coal, the pit price of coal is directly related to the
realm and therefore the pit price is the focus of this article. Historical tr ends show that the pit price of
raw coal has experienced a number of upward and downward movements, and the fluctuations up and
down have been dramatic.

Data processing and coal price forecasting. By collecting raw coal pithead prices over a ten-year period
from July 2010 to June 2020, the daily raw coal prices were chosen to obtain monthly average prices to
better reduce the error in price prediction. A total of 120 monthly coal price data were selected, and the
data set was sliced in order, with the first 100 sets of data used as the training set and a 5-fold
cross-validation to ensure model training accuracy, and the last 20 sets of data used as the test set. The
training set was used to train each model, and the test set was used to test and update the parameters of
the model.
The sparrow population was selected as 30, and the maximum number of iterations was set as 50.
The ISSA-LSSVR prediction model was used to predict the coal price time series according to the flow in
Figure 3. The sparrow algorithm operation results showed that: The optimal penalty factor C=1.9524, the
kernel function parameter g=0.1, and the deviation term is 0.054, that is, only the data exceeding the
existing coal price by ±5.4% is included in the error. The coal price prediction results are shown in Figure
5 and Figure 6.
500 450
Actual data
predicted value
400
400
350

predicted value
300
Actual data

300

250

200
200

150
100
100

0 50
2010 2012 2014 2016 2018 2020
years

Figure 5. ISSA-SVR model prediction results

50

48

46
Adaptation degree

44

42

40

38

36

34

0 10 20 30 40 50
Iteration times

Figure 6. Plot of adaptation with number of iterations

It can be seen that the Sparrow algorithm can approach the optimal fitness value at three iterations,
which illustrates the effectiveness of the ISSA algorithm; the sample fit of the training set is high, with R2
being 0.94247; the accuracy of the test set is poor, with R2 being 0.18223, indicating that the algorithm in
the article is less capable of handling abnormal fluctuations in the data, but the prediction results can
analyse the trend of coal price fluctuations and have certain application effects.

Coal price forecasting model accuracy test and analysis of forecasting results. To illustrate the effect
of the algorithm proposed in the article, a traditional time series forecasting model is selected for
comparative analysis. The autoregressive moving average model (ARIMA) is more widely used in the
forecasting of price time series in recent years. The model is mainly controlled by three indicators, such
as the number of autoregressive terms p, the number of moving average terms q and the number of
differences d. That is, the ARIMA (p,d,q) model takes the form of

 ( L )(1 − L ) X t =  ( L )  t
d
（16）
 Where:  ( L ) and  ( L ) are stationary lag operators of order p and order q respectively; L is the

lag operator, i.e. LYt = Yt −1 . {εt} is the white noise sequence; d is the difference parameter; c is a constant;
p, d and q are non-negative integers.
In general, the modeling process of price series predicted by ARIMA model is shown in Figure 7.
Obtain historical coal
price data

N differential
stationarity test processing

Whether it conforms to Y End of

white noise analysis

N
Initial determination of model type
based on ACF and PACF

model parameter
estimation

Calculate residual Statistics

in Computational Models

Testing for residuals

N
Is it appropriate

used to predict

Figure 7. The ARIMA model prediction flow chart

The stability analysis showed that the original data was a non-time smooth series, which was
differenced and the 1st order differenced series was a smooth time series. The model parameters were
chosen as ARIMA(2,1,0). the prediction results of the ARIMA model are shown in Figure 8.
500 320
Actual data
Predicted value

400 300
Predicted value

280
Actual data

300

260
200

240
100

220

0
2010 2012 2014 2016 2018 2020
years

Figure 8. Coal price ARIMA model prediction results

Two indexes including root mean square error (RMSE) and coefficient of determination (R2) were
used to evaluate the validity of the prediction model. The formula is as follows.
N

 ( yˆ − y )
2
RMSE = i i （17）
i =1
  ( yi − yˆi )2
N
2 i =1
R =1− （18）
 ( y − y )2
N
i =1 i

A comparison of the ISSA-LSSVR model and the ARIMA model predictions is shown in the table 1.
Table 1. Comparison of predictive model effects
Model categories Training set Test set

RMSE R2 RMSE R2

ISSA-LSSVR 26.7215 0.9425 50.2477 0.1822

ARIMA 108.3521 0.8648 117.5293 0.1236

Combined with Figure 5, Figure 8 and Table 1, it can be seen that both the training set and the test
set data show that the ISA-LSSVR model has good fitting accuracy. Meanwhile, due to the characteristics
of the SVR model, certain redundancy can be considered in the process of prediction, which conforms to
the idea of flexible optimization and optimizes the flexible interval to some extent.

Optimal design of realm opencast mine under coal price

fluctuations

Uncertainty risk nesting properties. A disadvantage of traditional realm optimisation methods is that
the parameters used in the calculation of the economic value of a block are fixed values and do not take
into account the effects of uncertainty. In order to choose an appropriate risk measure for the above
problem, it is necessary to define the ideal characteristics of the pit generated by the risk measure ρα
when varying the risk aversion α. The idea of nesting can be applied to generate a set of nested pits
containing ordered levels of risk, provided that the risk is managed.
Combining the constraints on constant normalisation and translation consistency to be satisfied by
the risk metric with the nesting idea, the risk nesting property can be derived for the open-air optimal
design process [28].
(1)Risk nestedness
Assuming that the blocks remain independently distributed within the mine area, as the parameter α
increases, the risk measure ρα of the decision process increases and the pit generated using the lower risk
aversion level α1 should be fully contained within the pit generated using the higher risk aversion level α2.
(2)Additive consistency
Suppose the random variable Z is independent of the intrinsic random variables X and Y, the variable
ρα( ) represents the risk measure at risk aversion α. If （X） （Y）→ （X +Z） （Y +Z）, then ρα( ) is said to
satisfy additive consistency. That is, the addition of a certain number of blocks to the pit at different
stages cannot change its original risk measure.
As shown in Figure 9, when the production of an open pit mine is affected by uncertainties, the
option of producing according to the original design boundary is questioned. the positive or negative
value of α can be used to reflect the different risks faced by the open pit mine: when α<0, the tendency is
to take the risk and the open pit boundary needs to be expanded to reduce the company's lost profit; when
α>0, the tendency is to avoid the risk and the open pit boundary needs to be reduced to When α>0, the
tendency is to be risk-averse and the open pit boundary should be reduced to minimise the loss of profit.
Combined with the SVM method, an interval band with a width of 2ε is constructed with the preset coal
price as the middle line: when the actual coal value is in the interval band, no loss is calculated, i.e. the
value of α is considered not to change; when the actual coal price is outside the interval band, the value of
α is considered to change, and the realm needs to be adjusted at this time.
Coal prices in the
redundancy interval
α=0 coal seam roof coal seam floor

Coal price exceeds

lower limit of
redundancy interval
Coal price exceeds α＞0
upper limit of
Risk Avoidance Boundary Line
redundancy
interval Original design boundary line
α＜0
Risk-taking Boundary Line

Figure 9. Schematic diagram of nested pits for the boundary optimization problem of open pit mine

Optimization and adjustment of coal limit under the fluctuation of coal price. In the process of the
original limit design of open-pit mine, the fixed coal price index is used as the basis for delineating the
limit, which lacks the adaptability to the fluctuation of coal price. Usually, the greater the difference
between the fixed coal price index and the actual data of coal price, the greater the risk the enterprise will
bear.
ISSA algorithm shows that when the fluctuation value of coal price exceeds the existing price by
±5.4%, the limit needs to be adjusted. In this case, the corresponding relationship between the final
mining limit change and the coal price fluctuation is shown in Figure 10.
500

400
Predicted value

300

200

100

0
2010 2012 2014 2016 2018 2020
years

(a) Coal price time series fluctuation trends

+ + + +
Advancing direction

+ Post-adjustment realm
+
+ Original design realm
+

(b) Trends in boundary changes in open pit mines over the corresponding cycle
Figure 10. Comparison of final state adjustment of open pit mines and coal price volatility
The trend of the coal price series shows that the raw coal price has gone up and down twice, with an
average price of 285.85 yuan, and the coal price is highly volatile. Dynamic adjustment of the final
mining realm of the open pit based on coal price fluctuation trends can effectively address the risks
arising from deviations of pre-set profit targets from actual scenarios and enhance corporate efficiency.
6. Case study

Topology of a small inclined coal seam opencast mine. To illustrate the effectiveness of the method in
the article for solving open pit mine realm optimisation problems under uncertainty, an idealised small
open pit coal mine was selected for the study. To ensure the validity of the model, it is assumed that the
advancement of this opencast coal mine remains consistent across the range of years. The
two-dimensional topology of this small inclined coal seam opencast mine is shown in Figure 11.

Outer dumping redd

Inner dumping redd
mining boundary
coal seam
b
Figure 11. Two-dimensional topology structure of a small inclined coal seam open pit mine
It is assumed that the coal seam dip Angle and final slope Angle are both 45°. At the same time, the
inner and outer stripping are distinguished, and the constraint of block extraction priority is given. It is
assumed that the final mining limit of the open-pit mine is variable when the external environment
changes, that is, when the coal price exceeds the preset value, the mining depth can be appropriately
increased to expand the open-pit mine limit and raw coal output, and vice versa.
For all blocks within the mining scope, mining cost cb is assigned. In addition, given the priority
constraint of each block, it follows the constraint of maximum side slope Angle. In order to simplify the
description process, pb is taken as the random variable representing the profit of the block. It is assumed

that the cost of mining unit volume of the block is cbe , the unit transportation cost is cbt , the unit

post-processing cost of the block is cbp , the internal and external stripping distances are taken as Ln and

Lw respectively, and the raw coal distances are taken as Lm. The extracted block cost vb can be calculated
by the following formula:

( )
cbe + cbt  Lm + cbp − rb pb （Ore block）


vb =  cbe + cbt  L（
n Inner row waste block） （19）
 e
 cb + cb  L（
t
w Outer row waste block）

Where: pb - profit of treated unit block (selling price of raw coal under deterministic condition);
rb  0 compensates for other factors that affect block profits.
Let P  B  B be the priority constraint set, that is, for (b,b’)∈P, block b’ must be mined before
block b, and then the medium and long term boundary optimization problem is expressed with cost
minimization:
 UP = min  (cbe xbe + cbt xbn  Ln + cbt xbw  Lw
x e, x p bB

+cbt (
 Lm ) +  cbp − rb pb xbp
bB
)
s.t. （20）
xbe  xbe, ,  ( b, b )  P,
,

xbe , xbp , xbn , xbw  0,1, b  B,

Where: xbe  xbe, is the block priority constraint; xbp is post-processing decision; xbe is the

mining decision; xbn is internal scheduling decision; xbn is the outgoing decision; xbp , xbe , xbn , xbw

and are binary random variables. The value can be 0 or 1.

Limit adjustment of small open-pit mine under coal price fluctuation. Generally, the range of coal
price change is small, so it is assumed that the range of coal price change is the current coal price ±50%.
At the same time, in order to ensure the continuity of the project, it is assumed that the change of the
actual coal price caused by the change of the boundary is not more than 20% of the existing coal price,
and different coal price fluctuation ranges correspond to different α values: (1) When the change of coal
price is in the range of (-5.4%,5.4%), α is considered to be 0; (2) When the coal price changes are at
[-50%,-5.4%] and [5.4%,50%], the data are normalized and the corresponding α value is calculated, that
is, α value of ±5.4% corresponds to 0, α value of 50% corresponds to -1, α value of -50% corresponds to
1. The part exceeding ±50% shall be calculated as ±50%.
The fluctuation cycle of coal price in Section 3.2 is selected as the object of this study, and Fig.10 (a)
is referred to. The coal price presents a combined trend of change, and the adjustment of the boundary
should follow the change of the coal price. However, considering the range of the adjustment of the
boundary, the length of the working line should be as close as possible to the length of the working line
calculated from the actual coal price, as shown in Figure 12.

0 +0.1 +0.2
-0.1 0 +0.1 +0.2
-0.2 -0.1 0 +0.1 +0.2
-0.2 -0.1 0 +0.1 +0.2
-0.2 -0.1 0
-0.2

Figure 12. Schematic diagram of two-dimensional boundary adjustment under risk condition
As shown in Figure 12: (1) When α=0, it represents risk neutrality, that is, there is no deviation
between the preset coal price index and the actual coal price, and there is no need to adjust the level at
this time; (2) When α is +0.1 and +0.2, it means that the preset coal price of open-pit mine is higher than
the actual coal price, and the enterprise needs to bear the risk of loss. In this case, it is necessary to reduce
the open-pit limit. (3) When α is -0.1 and -0.2, it means that the preset coal price of open-pit mine is
lower than the actual coal price. In this case, it is necessary to expand the open-pit mine limit.
It can be seen that, for open-pit mine of inclined coal seam, affected by the occurrence state of coal
seam, depth reduction should be considered when the boundary is expanded, and the difficulty of
boundary expansion is obviously higher than that of boundary reduction. Therefore, only when the coal
price is higher than the preset coal price for a long period of time, the boundary will be expanded. That is,
for the final boundary plan, the amplitude and frequency of boundary contraction are significantly higher
than that of expansion, as shown in Fig 10 (b).
It is assumed that the small open pit coal mine is in a partial internal discharge period from 2010 to
2022, that is, a small part of the stripping is transported to the outer dump, and most of the stripping is
transported to the inner dump for discharge. To simplify the description process, it is assumed that the top
flat strip is transported to the outer dump and the remaining flat strip is internally discharged. The block
in the mining limit contains three directions: inner and outer dump and coal storage bin. Given the main
economic indicators of the production link of the open-pit mine, see Table 2.
Table 2. Main economic indicators in the production link
name of index unit numerical value
Cost of extraction yuan·m-3 10
Cost of transportation yuan·m-3·km-1 2
Post-processing cost yuan·m-3 5
Internal drainage distance km 2
Outbound running distance km 4
Raw coal transport distance km 3
Preset coal profit yuan·m-3 -60
Recovery of resources % 95
The block model profile of a small open-pit mine is shown in Figure 13. The blocks labeled "1", "2"
and "3" are respectively transported to the outer dump, the inner dump and the coal storage bin, assuming
that the volume of the blocks is 1×1×1m. Below, the profits in different mining limits are described in
terms of different values of α.
¦=
Á0 Á+0.1 ¦ Á
¦= =+0.2

¦=
Á-0.1
1 1 1 1 1 1 1 1 1 3 3
¦=
Á-0.2
2 2 2 2 2 2 2 3 3
2 2 2 2 2 3 3
2 2 2 3 3
2 3 3
3
Figure 13. Schematic diagram of the block model section of a small open-pit mine
(1)α=0
At this time, there is no deviation between the actual coal price and the preset coal price, that is, the
risk is neutral. According to Equation (20), the net profit in the mine
W0=9×(10+2×4)+16×(10+2×2)+11×(10+2×3+5-60×0.95)=-10 yuan.
(2)α<0
Considering that the expansion of the limit should be accompanied by the depth reduction of the
open-pit mine, the limit can be expanded only when the actual coal price exceeds 20% of the preset coal
price, i.e., α=-0.2. Take profit of raw coal is 67.6 yuan, according to the equation (20) mine in net profit
calculated W-0.2=9×(10+2×4)+24×(10+2×2)+12×(10+2×3+5-67.6×0.95)=-20.4 yuan.
(3)α>0
When the actual coal price is 10% lower than the preset coal price, α=+0.1, then the level can be
reduced. Take profit of raw coal is 56.8 yuan, according to the equation (20) mine in net profit calculated
W+0.1=8×(10+2×4)+12×(10+2×2)+10×(10+2×3+5-56.8×0.95)=-12 yuan.
It can be seen that when the raw coal price increases by 20% and decreases by 10%, the adjustment
of the open-pit limit can increase the net profit of the pit by about 104% and 20%, and at the same time
ensure that the risk is under control.
Coal price forecast period dynamic adjustment effect test. As mentioned above, different boundary
adjustment thresholds can be adopted according to the construction and development period of the
open-pit mine, that is, the value of α is a dynamic process. For the convenience of narration, this article
assumes that the working line length of the small open-pit mine does not change greatly. The coal price
fluctuation trend in Figure 10 (a) was used as the basis for the boundary adjustment, and the boundary
adjustment period was set as 10 years. The comparison between the actual coal price and the preset coal
price in each year during the adjustment period is shown in Table 3.
Table 3. Comparison of actual and preset coal price during the adjustment period
Demand
Actual Price of Coal price
The range of coal for
Year coal price coal used adjustment α value
price change boundary
(RMB) (Yuan) range
adjustment
2009 — 285.85 — — — —
2010 340.93 340.93 16.16% 16.16% -0.24 Expansion
2011 408.23 408.23 16.48% 16.48% -0.25 Reduce
2012 332.99 332.99 -18.43% -18.43% 0.29 Reduce
2013 237.07 266.39 -28.80% -20.00% 0.52 Reduce
2014 172.07 213.11 -35.40% -20.00% 0.67 Reduce
2015 110.27 170.49 -48.26% -20.00% 0.96 Reduce
2016 190.03 190.03 11.46% 11.46% -0.14 Expansion
2017 353.40 228.04 85.97% 20.00% -1.00 Expansion
2018 347.09 273.64 52.21% 20.00% -1.00 Expansion
2019 313.28 313.28 14.48% 14.48% -0.20 Expansion
2020 286.01 286.01 -8.70% -8.70% 0.07 Reduce
As can be seen from Table 3, when the degree of achievement cannot meet the demand at a certain
period during the adjustment period, it is necessary to adjust the open-pit mine limit in time to meet the
risk minimization demand. Figure 14 shows the comparison between stage coal price and actual raw coal
price in the process of multi-step boundary optimization method. The coal price used in each stage fits the
trend of actual coal price well. Except for the period of abnormal change of coal price, the coal price used
in most stages is within the range of ±20% of actual coal price.
450 450
Actual coal price
Operational coal price
400
400

350
Operational coal price

350
Actual coal price

300

300
250

200 250

150
200

100
150
2008 2010 2012 2014 2016 2018 2020
Coal price forecast period

Figure 14. Comparison between the used and actual coal price in multi-step boundary optimization design process
Conclusion

In this article, by introducing the flexible idea into the coal price prediction process, LSSVR
prediction model is proposed to predict the coal price time series. Considering the given regularization
term C and kernel function parameter g of traditional LSSVR, the prediction process is too subjective.
The ISSA algorithm was used to optimize the above two parameters in the ISSA-LSSVR model. By
taking the two parameters as the optimal position of the sparrow population, the optimal C and g values
were obtained as 1.9524 and 1.00 respectively. The accuracy test of the model shows that the calculation
accuracy and error of the proposed algorithm are improved compared with the traditional ARIMA method,
and the predicted results have a high degree of fitting with the actual values. Combined with the risk
nesting property in the process of open-air optimization design, the mathematical formula of boundary
optimization problem is given in terms of cost minimization. By combining the coal price prediction
results with the boundary optimization problem, the final mining boundary of the open-pit mine is
dynamically adjusted according to the coal price fluctuation trend, and the risk of the preset profit index
deviating from the actual scene is effectively dealt with. Taking an idealized small open-pit coal mine as
an example, different risk aversion degrees are given according to different coal price fluctuations, and
then the open-pit mine limits are adjusted. The results show that the use of coal price in each stage is a
good fit for the trend of actual coal price, except for the period of abnormal change of coal price, the use
of coal price in most stages is within the range of ±20% of the actual coal price.

Data availability

The data used to support the findings of this study are available from the corresponding author upon
request.

References

[1] Cohen, M. W. & Coelho, V. N. Open-Pit Mining Operational Planning using Multi Agent Systems.
Procedia Computer Science 192(2021): 1677-1686 (2021).
[2] Gu, X.W., Qing, W. & Shu, G. Dynamic phase-mining optimization in open-pit metal mines.
Transactions of Nonferrous Metals Society of China 20(10): 1974-1980 (2010).
[3] Fathollahzadeh, K., Asad, M. W., Mardaneh, E. & Cigla M. Review of solution methodologies for
open pit mine production scheduling problem. International Journal of Mining, Reclamation and
Environment 35(8): 564-599 (2021).
[4] Jélvez, E., Morales, N., Nancel-Penard, P., Peypouquet, J. & Reyes, P. Aggregation heuristic for
the open-pit block scheduling problem. European Journal of Operational Research 249(3):
1169-1177 (2016).
[5] Suparno, F., Paithankar, A. & Chatterjee, S. Developing risk assessment of push-back designs for
an Indonesian coal mine under price uncertainty. AIP Conference Proceedings 2245(1): 080002
(2020).
[6] Asad, M. W. & Dimitrakopoulos, R. Performance evaluation of a new stochastic network flow
approach to optimal open pit mine design-application at a gold mine. Journal of the Southern
African Institute of Mining and Metallurgy 112(7): 649-655 (2012).
[7] Zhao, Z., Zhu, J. & Xia, B. Multi-fractal fluctuation features of thermal power coal price in China.
Energy 117: 10-18 (2016).
[8] Zhang, K., Cao, H., Thé, J. & Yu, H. A hybrid model for multi-step coal price forecasting using
decomposition technique and deep learning algorithms. Applied Energy 306: 118011 (2022).
[9] Lütkepohl, H. Vector autoregressive models. Handbook of Research Methods and Applications in
Empirical Macroeconomics 2013: 139-164 (2013).
[10] Box, G. E., Jenkins, G. M., Reinsel, G. C. & Ljung, G. M. Time series analysis: forecasting and
control. John Wiley & Sons (2015).
[11] Andersen, T., Bollerslev, T. & Hadi, A. ARCH and GARCH models. John Wiley & Sons (2014).
[12] Bonita, O. & Muflikhah, L. Comparison of Gaussian and ANOVA Kernel in Support Vector
Regression for Predicting Coal Price. 2018 International Conference on Sustainable Information
Engineering and Technology (SIET) IEEE 2018: 147-150 (2018).
[13] Herrera, G. P., Constantino, M., Tabak, B. M., Pistori, H., Su, J. J. & Naranpanawa, A. Long-term
forecast of energy commodities price using machine learning. Energy 179: 214-221 (2019).
https://doi.org/10.1016/j.energy.2019.04.077.
[14] Panella, M., Barcellona, F. D. & Ecclesia, R. L. Forecasting Energy Commodity Prices Usi
ng Neural Networks. Advances in Decision Sciences 2012: 1–26 (2012). https://doi.org/10.1
155/2012/289810.
[15] Liu, Z., Zhou, P., Chen, G. & Guo, L. Evaluating a coupled discrete wavelet transform and support
vector regression for daily and monthly streamflow forecasting. Journal of Hydrology 519:
28220-2831 (2014).
[16] Niu, W. J., Feng, Z. K., Xu, Y. S., Feng, B. F. & Min, Y. W. Improving prediction accuracy of
hydrologic time series by least-squares support vector machine using decomposition reconstruction
and swarm intelligence. Journal of Hydrologic Engineering 26 (9): 04021030 (2021).
[17] Li, B. J., Sun, G. L., Li, Y. P., Zhang, X. L. & Huang, X. D. A hybrid model of variational mode
decomposition and sparrow search algorithm-based least square support vector machine for
monthly runoff forecasting. Water Supply (2022).
[18] Bhagat, S. K., Tiyasha, T., Awadh, S. M., Tung, T. M., Jawad, A. H. & Yaseen, Z. M. Prediction of
sediment heavy metal at the Australian Bays using newly developed hybrid artificial intelligence
models. Environmental Pollution 268: 115663 (2021).
[19] Su, X., He, X., Zhang, G., Chen, Y. & Li, K. Research on SVR Water Quality Prediction Model
Based on Improved Sparrow Search Algorithm. Computational Intelligence and Neuroscience
(2022).
[20] Ehteram, M., Sharafati, A., Asadollah, S. B. H. S. & Neshat, A. Estimating the transient storage
parameters for pollution modeling in small streams: a comparison of newly developed hybrid
optimization algorithms. Environmental Monitoring and Assessment 193(8): 1-16 (2021).
[21] Xue J. & Shen B. A novel swarm intelligence optimization approach: sparrow search algorithm.
Systems Science & Control Engineering 8(1): 22-34 (2020).
[22] Lv, X., Mu, X. D., Zhang, J. & Wang Z. Chaos sparrow search optimization algorithm. Journal of
Beijing University of Aeronautics and Astronautics 47(8): 1-10 (2021).
[23] Mao, Q. H., Zhang, Q. Q., Mao, C. C. & Park, J. X. Hybrid sine cosine algorithm and sparrow
algorithm for Lévy flight. Journal of Shanxi University (Natural Science Edition) 44(6):
1086-1091 (2021).
[24] Tang, A. & Han, T. UAV trajectory planning method based on chaotic sparrow search algorithm.
Computer application 5(7): 1-11 (2021).
[25] HARRIS D., CHRISTOPHER J. C. B. & LINDA K. Support vector regression machine. Advances
in Neural Information Processing Systems 9 (NIPS). 155-161 (1997).
[26] Jelvez, E., Morales, N. & Ortiz, J. M. Stochastic Final Pit Limits: An Efficient Frontier Analysis
under Geological Uncertainty in the Open-Pit Mining Industry. Mathematics 10(1): 100 (2021).
[27] Morales, N., Seguel, S., Cáceres, A., Jélvez, E. & Alarcón, M. Incorporation of geometallur-gical
attributes and geological uncertainty into long-term open-pit mine planning. Minerals 9(2): 108
(2019).
[28] Groeneveld, B. & Topal, E. Flexible open-pit mine design under uncertainty. Journal of Mining
Science 47(2): 212-226 (2011).
[29] Canessa, G., Moreno, E. & Pagnoncelli, B. K. The risk-averse ultimate pit problem. Optimization
and Engineering 22(4): 2655-2678 (2021).
[30] Jiang, S., Yang, C., Guo, J. & Ding, Z. ARIMA forecasting of China’s coal consumption, price and
investment by 2030. Energy Sources, Part B: Economics, Planning, and Policy 13(3): 190-195
(2018).
[31] Lasheras, F. S., de Cos Juez, F. J., Sánchez, A. S., Krzemień, A. & Fernández, P. R. Forecasting the
COMEX copper spot price by means of neural networks and ARIMA models. Resources Policy 45:
37-43 (2015).
[32] Shishvan, M. S. & Sattarvand, J. Long term production planning of open pit mines by ant colony
optimization. European Journal of Operational Research 240(3), 825-836 (2015).
[33] Choudhury, S. & Chatterjee, S. Pit optimisation and life of mine scheduling for a tenement in the
Central African Copperbelt. International Journal of Mining, Reclamation and Environment 28(3),
200-213 (2014).
[34] Jélvez, E., Morales, N., Nancel-Penard, P., Peypouquet, J. & Reyes, P. Aggregation heuristic for
the open-pit block scheduling problem. European Journal of Operational Research 249(3),
1169-1177 (2016).