Professional Documents
Culture Documents
https://doi.org/10.1007/s11081-019-09479-6
RESEARCH ARTICLE
Abstract
We present a mixed-integer programming model for solving the long-term planning
problem of an underground mine. This model, which establishes the sequence of
mining for a horizon of 20 years, determines which lens of the geological model
will be mined and in what order, while respecting the operational constraints. For
each lens to be mined, a specific cut-off grade has to be selected to maximize the net
present value. The choice of a cut-off grade affects the volume and the average grade
of each lens, which increases the size of problems to be solved. To reduce the com-
putation time, different acceleration strategies and a Fix-and-Optimize heuristic are
proposed. Computational experiments on instances of different sizes are performed
to (1) assess the quality of the solution found by each method and (2) present the
impact of the variable cut-off grade.
* Rafael Martinelli
martinelli@puc‑rio.br
Jean Collard
jean.collard@polymtl.ca
Michel Gamache
michel.gamache@polymtl.ca
1
Departamento de Engenharia Industrial, Pontifícia Universidade Católica do Rio de Janeiro
(PUC-Rio), Rua Marquês de São Vicente, 255 – Gávea, Rio de Janeiro, RJ 22451‑900, Brazil
2
Department of Mathematics and Industrial Engineering, École Polytechnique de Montréal,
C.P. 6079 succ. Centre‑Ville, Montréal, QC H3C 3A7, Canada
3
Groupe d’Études et de Recherche en Analyse des Décisions (GERAD), HEC Montréal 3000
chemin de la Côte-Sainte-Catherine, Montréal, QC H3T 2A7, Canada
13
Vol.:(0123456789)
R. Martinelli et al.
Raglan Mine, owned by Glencore, is located in the Nunavik region in the north of
the province of Quebec (62nd parallel) about 1800 km from Montreal. The nickel
ore resources of the mine are grouped into several small deposits of elongated
and finite shapes called lenses. The lenses are distributed over the entire property
(which stretches about 70 km from east to west) and have different tonnages and
grades. Due to the lens dispersion on the mining property, more than one mine
is required to extract the ore. Lenses assigned to a mine are interconnected by a
main ramp and a network of drifts (horizontal tunnels) and share the same infra-
structure (such as shelters and ventilation systems).
The long-term or strategic planning in mining operations consists of schedul-
ing the mining activities to maximize the net present value of profits over a plan-
ning horizon, which is estimated based on the geological reserves. This problem
is often called the Life Of Mine (LOM) planning by mining engineers. At Raglan
Mine, the long-term planning horizon extends over 15–20 years. The long-term
planning involves several strategic decisions including, among others, the open-
ing and closing of a mine, the choice of cut-off grade (to be explained later), the
decision whether or not to exploit a lens (and, if so, when it should be exploited),
the determination of the rate of production of lenses, the start of ramp and drift
excavation, etc.
The cut-off grade is a concept used by mining engineers to determine if a block
of material should be considered ore (so exploited and sent to the ore process-
ing plant) or waste (rock without economic value that preferably should remain
in place). To achieve this ranking, geologists divide the deposit surrounding
each lens in cubic blocks of 125 m 3 (i.e., 5 m × 5 m × 5 m). Based on the data
of exploration drilling, a geostatistical approach is used to estimate the various
geomechanical and geochemical characteristics of each block, among them the
ore grade. After that, all blocks with a grade higher than or equal to the cut-
off grade will be considered as a block of ore, while other blocks will be waste
blocks. For the Raglan Mine deposit, given that the blocks of higher grade are at
the center and bottom of lenses and those of lower grade are at the periphery, the
choice of the cut-off grade will impact the morphology of lenses. When the cut-
off grade increases, the average grade of the lens also increases, but the tonnage
of the latter decreases.
Usually, the selection of the cut-off grade and strategic planning are performed
sequentially. The value of cut-off grade is first selected, then the mine design and
the long-term planning are carried out according to this cut-off grade. In this
paper, we propose a mixed-integer programming (MIP) model for solving the
strategic planning at Raglan Mine. This model takes into account all the opera-
tional constraints of the mine deemed essential for long-term planning. In addi-
tion, the MIP model lets the optimizer determine the best cut-off grade for each
lens to maximize the net present value of profit.
This paper contributes to fill a methodological gap as (1) it introduces the
notion of flexible and independent cut-off grade for each lens; (2) it takes into
13
Strategic planning of an underground mine with variable cut‑off…
The operations of underground mines can be divided into two groups of activities:
development activities and production activities. The details of the activities in each
group vary significantly from one underground mine to another. In this paper, we
describe only the activities encountered at the Raglan Mine. The explanation of
these activities is essential to better understand the description of the mathematical
formulation, which will follow in the next section.
To access the lenses, engineers must excavate ramps, also called inclines, and
drifts, which are horizontal cuts that allow access to the lenses from the ramp. These
excavation activities are carried out in areas of waste rock, yielding no income, and
they generate considerable expenses. However, these development activities are
essential to exploit the lenses. Development activities are divided into two catego-
ries: the development of common infrastructure (CAPEX, which stands for CAPital
EXpenditure) and operational development (OPEX, which stands for OPerational
EXpenditure). CAPEX activities consist of infrastructure that will be used through-
out the life of the mine, while OPEX activities consist of the development of drifts
giving access to lenses from the main ramp. If the ore of a lens is not extracted,
the OPEX development associated with this lens will not be made. The number
of meters of OPEX development done for a lens depends on the size of the lens.
Indeed, a lens with a high tonnage requires more OPEX development. Both types of
development are shown in Fig. 1.
Several mining methods can be encountered in underground mines. The choice
of the mining method depends mainly on the mechanical properties of the bedrock.
At Raglan Mine, two mining methods are used to extract the ore from lenses: the cut
13
R. Martinelli et al.
Surface
1
mp
Ra
Lens 1 drift
drift
CAPEX
Lens 2
drift Cuts for long hole
Ra
mp
2
OPEX
m
Ra
Lens 3 Ra
mp
4
Fig. 1 Mining operations
and fill method and the long hole method. Cut and fill mining is a selective method
of mining in which horizontal cuts (slices of ore) are removed advancing upwards.
Upon completion, the cut is filled back to the access ramp with backfill mate-
rial, which is often composed of waste material excavated from ramps and drifts.
Another drift is driven on top of the filled cut. This process continues until the top of
the stope is reached. The method is shown at the bottom of lens 2 in Fig. 1.
The long hole method consists first of drilling horizontal cuts through the ore
(see cuts for long hole in Fig. 1). These cuts separate the orebody into sublevels, and
each sublevel is divided into stopes. These cuts, which are used to enable the miner
to blast off vertical slabs of ore, are similar to those done by the cut and fill method,
and, for this reason, they are considered in the MIP model as being part of the cut
and fill activities. Using these cuts, the miner drills a set of parallel vertical holes,
with a height equal to one sublevel, and fills them with explosives. Blasted slices
of ore run into an open void within the open stope and are mucked out with load-
ing equipment to the surface using ramps. Stopes are typically backfilled after they
are mined out. Between stopes, ore sections are set aside for pillars to support the
hanging wall. These pillars are generally shaped as vertical beams across the ore-
body. This method of mining is very cheap compared to cut and fill mining because
it facilitates the extraction of high volumes of ore in a shorter time and with fewer
pieces of equipment.
13
Strategic planning of an underground mine with variable cut‑off…
At Raglan Mine, the cut and fill method is used at the bottom of the lens, when
the dip of the lens is low and prevents the use of the long hole method. This tech-
nique operates a small proportion of each lens. The remainder of the lens is operated
by long hole method. The proportion of the two methods varies from one lens to
another. It depends on the characteristics of the lens and also on the selected cut-off
grade. It is essential to know the proportion of each lens to be operated by each min-
ing method given that their production rates differ significantly: the long hole min-
ing method has a much higher production rate than the cut and fill mining method.
Mining seldom recovers all resources present in an ore deposit. The amount
of ore extracted from a deposit is referred to as the mining recovery rate and is
expressed as a percentage. In addition, a certain amount of waste is usually mixed in
with the ore during mining. A dilution factor is used to take into account this phe-
nomenon. Finally, the amount of nickel in the ore cannot be recovered at 100% at the
processing plant. A recovery factor at the processing plant should also be considered
to more accurately estimate the amount of extracted nickel. All these factors were
provided by the engineering team at Raglan Mine.
When a stope is excavated, the void left by the excavation creates pressure on
the rock that surrounds it, and this may cause subsidence. To avoid subsidence,
the stopes must be backfilled, and engineers use waste rock from the excavation of
ramps and drifts as backfill material. During a year, if more waste is extracted than
ore, the surplus that is not used for backfilling stopes must be transported to the
surface. Conversely, if more ore was extracted in a year, waste material stored on
the surface will have to be descended into the mine to compensate for the lack of
backfill material. Both operations raise operational costs, and these costs must be
considered in the MIP model for each year of the planning.
Mining operations involve some precedence constraints between activities. For
example, in Fig. 1, ramp 1 must be done before ramp 2, and ramp 2 before ramp 3,
etc. The OPEX development of lens 1 can be done only if ramp 1 is constructed.
Moreover, the ore extraction by the cut and fill mining method can begin only if a
proportion of OPEX was performed. As shown in Fig. 1, all OPEX development
does not need to be performed before starting the extraction of ore by the cut and fill
method. Similarly, the extraction of ore by the long hole mining method may begin
if the development of cuts for the long hole was made. To allow the overlapping
of certain mining activities with certain development activities, in the MIP model,
OPEX development, cut and fill, and long hole mining activities of each lens were
split into two parts. This separation reflects the percentage of OPEX development
needed before starting mining by the cut and fill method, and similarly, the percent-
age of the cut and fill method (which, as stated before, includes the long hole cuts)
that must be done before beginning long hole mining activities. However, there is
no precedence constraints between the lens. For example in Fig. 1, lens 2 can be
extracted before lens 1.
Figure 2 presents the precedence relationship between all activities of a mine m
using a graph denoted Gm = (Nm , Am ) . In this graph, each node r𝓁m represents the
ramp segment whereby development that gives access to lens 𝓁 will be made, while
nodes o1𝓁,m , o2𝓁,m , c1𝓁,m , c2𝓁,m , h1𝓁,m , and h2𝓁,m represent the first and second parts of the
OPEX development, the first and second parts of the cut and fill method, and the
13
R. Martinelli et al.
lens 1,m
c11,m c21,m
lens 2,m
lens k,m
first and second parts of the long hole mining method associated with lens 𝓁 , respec-
tively. Each arc (i, j) ∈ Am indicates that activity i must be done before activity j.
Over the last two decades, the use of Operations Research techniques for under-
ground mine planning has increased, but it is still more common in open-pit mines.
As mentioned in Newman et al. (2010), the fact that underground projects vary sig-
nificantly from one operation to another, because the set of constraints and the min-
ing method are proper to a deposit, explains the singularity of each underground
project. Most models in the literature for medium- and long-term planning in under-
ground mine are based on integer linear programming approaches. Due to the com-
plexity of solving this problem optimally, most of them must either relax some con-
straints (especially those on the integrality of the binary variables), aggregate tasks
or group them to reduce the number of variables and constraints, or solve the prob-
lem sequentially.
In the literature on long-term planning in underground mines, two groups of
problems can be distinguished. The first group of problems deals with a shorter time
horizon (between 3 and 7 years) than the Life of Mine.
Dagdelen et al. (2002) work on a seven-year horizon, and since the deviation
between the target and the production is minimized for each month, the number of
variables is very high. To reduce its size, the problem is divided by year, and the
seven problems are considered independently of each other.
Kuchta et al. (2004) describe the use of mixed-integer programming to schedule
the operations of an iron ore underground mine in Sweden over a five-year hori-
zon. The objective function consists of minimizing each month the deviation from
specified production levels. Using preprocessing and judicious choice of variable
13
Strategic planning of an underground mine with variable cut‑off…
definition (which permits to aggregate blocks), they can develop a tractable solution
approach. The presence of lens in the geological model at Raglan makes it more
challenging to do such aggregations.
Martinez and Newman (2011) minimize the difference between the monthly pro-
duction of the mine and the monthly targets over a planning horizon of three years.
The model also positions the equipment on their level of work. To reduce computa-
tional time, the authors have developed a heuristic decomposition method.
The second group of problems deals with Life of Mine planning, and their objec-
tive consists of maximizing the net present value of profits.
In their model, Rahal et al. (2003) optimize the sequence of a mine using a block
caving method. The deposit nature is used to create links between the sites, and a
relaxation of the linear program obtains a good approximation of the solution.
Smith et al. (2003) attempt to maximize the Net Present Value (NPV) of the
three mines of Mount Isa Mines. In this article, the cut-off is fixed beforehand. The
authors merge the existing 1500 sites in 34 zones. The purpose is to reduce the
number of variables and constraints. To reduce the computation times, the authors
stopped the solution process when the gap between the best solution and the best
node in the branch and bound tree is lower than 10%.
Sarin and West-Hansen (2005) propose a Benders’ decomposition approach to
solve a mixed binary programming model for the long-term planning of an under-
ground coal mine. The model proposed by Nehring and Topal (2007) has similari-
ties with the model of this paper. Indeed, the authors divide each work into sev-
eral tasks that must be performed sequentially. In addition, they maximize the net
present value of their deposit. However, the investments are not considered in their
paper, and the proposed example is too simple, which makes it unclear whether the
model is applicable to real situations.
In more recent papers, a trend is to implement resource-specific constraints. Ter-
blanche and Bley (2015) then propose a united formulation for the long-term plan-
ning problem with simpler notation for resources and a modified version that opti-
mizes concerning a block selection within the mine. Epstein et al. (2012) maximize
the net present value of the largest copper deposit in the world. Two mining methods
are used: open pit and underground by a block caving method. Some binary vari-
ables are relaxed, and a heuristic is used to round fractions, reducing the compu-
tational time. This model also tracks tonnes mined along the production network.
However, the authors do not seem to take account of OPEX and CAPEX operations
in their model. Nehring et al. (2012) present a mixed integer programming formula-
tion of the long-term planning model that assigns the different resources and equip-
ment needed for every site. The model assigns a binary variable to every activity to
be done in every stope, and a variation of this model uses a single binary variable for
all activities under a more restrictive hypothesis. Little et al. (2013) present a model
to optimize stopes shape in parallel with planning. Their results showed that consid-
ering both aspects increases the NPV.
Finally, some authors opt for a heuristic approach; for example, Riff et al. (2009)
propose a GRASP (Greedy randomized adaptive search procedure) meta-heuris-
tic to optimize the sequence of operations of an underground mine. The method
maximizes the NPV keeping all constraints and all variables. This approach is a
13
R. Martinelli et al.
considerable advantage over methods that generally require the ILP relaxation of
certain constraints and removal of variables to allow for resolution of the problem.
This model does not select the cut-off grade but could be adapted to take it into
account.
In all these papers, the cut-off grade is fixed to a specific value for the entire
deposit. Some authors, Horsley (2005) and Wang et al. (2008), have sought to opti-
mize the cut-off grade using a pre-established mining sequence. In the current paper,
we try to deal with both aspects simultaneously: looking for an optimal mining
sequence and determining the optimal cut-off grade for each lens. The mixed-integer
program presented joins the idea proposed by Smith and O’Rourke (2005), claiming
that when the cut-off grades are variable, the solution will explore the parts of the
mines with the highest NPV first, considering all the constraints imposed, with no
need to follow any fixed cut-off policy beforehand.
In the remainder of this section, we describe the mixed-integer program (MIP)
used to solve the problem of long-term planning. First, we define the sets, param-
eters, and variables used. Then, the objective function and the constraints of the
model are detailed.
Based on the description previously given, we can define sets and subsets per
mine for each type of activity, as shown in Table 1. In this table, a set G of cut-off
Table 1 List of sets
Set Definition
M Mines
T Periods
G Cut-off grades
Lm Lenses of mine m ∈ M
L ∪m∈M Lm
Rm Ramp segments (CAPEX) of mine m ∈ M , i.e. Rm = {r1,m , r2,m , … , r∣Lm ∣,m }
R ∪m∈M Rm
Om OPEX development segments of mine m ∈ M , i.e. Om = {o11,m , o21,m , … , o2∣L }
m ∣,m
O ∪m∈M Om
Dm Development activities of mine m ∈ M , i.e. Dm = Rm ∪ Om
Cm Cut and Fill activities of mine m ∈ M , i.e. Cm = {c11,m , c21,m , … , c2∣L }
m ∣,m
C ∪m∈M Cm
Hm Long Hole activities of mine m ∈ M , i.e. Hm = {h11,m , h21,m , … , h2∣L }
m ∣,m
H ∪m∈M Hm
Pm Production activities of mine m ∈ M , i.e. Pm = Cm ∪ Hm
A𝓁 Activities of lens 𝓁 ∈ L
Am Activities of mine m ∈ M , i.e. Am = Dm ∪ Pm
A ∪m∈M Am
13
Strategic planning of an underground mine with variable cut‑off…
grade values has been created. The lens cut-off grade choice has an impact on
both its volume and its average grade. Two non-linear functions usually represent
the relations between the cut-off grade and the volume and the average grade of
a lens. We could have approximated these functions by a set of tangent lines in
order to insert them into the linear programming model. However, since the infor-
mation provided by these functions represent estimations of volume and aver-
age grade, the accuracy provided by these curves does not allow a careful choice
of cut-off grade; which is why we opted for a set G of discrete values of cut-off
grades taken at regular intervals.
Table 2 shows the parameters used in the MIP model.
The factor f t = (1 + d)t , where d is the discount rate, is used to discount the
cash flows from the net present value. Furthermore, the selling price of nickel
St at period t, the treatment cost CP and the recovery factor F R at the processing
plant can be used to calculate the revenue per ton of metal as Rt = F R (St − CP ) .
Notice that the mining recovery rate and the dilution factor are not specified
because they are already included in the parameters Tag and Nag for a ∈ Pm for
each m ∈ M . Finally, the swelling coefficient of waste rock F S and the loading
factor of an excavated lens F L must be considered for backfilling, and the cost
Table 2 List of parameters
Name Definition
13
R. Martinelli et al.
Note that there is no need to define variables that represent the opening of mines
because each mine m ∈ M is always started by a single CAPEX development,
which is represented by the variable yr1m t.
Besides the binary variables, the MIP model also includes two groups of continu-
ous variables, which are defined in Table 3.
3.2 Objective function
The objective is to maximize the net present value of Raglan Mine over a fixed plan-
ning horizon. With this aim, the objective function is divided into several parts. Part 1
shows the income obtained from the sale of nickel. The amount of nickel reflects the
Table 3 Continuous variables
Name Definition
xagt Fraction (%) of mining activity a ∈ A , based on a cut-off grade g ∈ G , completed at time t ∈ T
wout
mt Volume ( m3 ) of waste material from mine m ∈ M to be stored at the surface at time t ∈ T
win
mt Volume ( m3 ) of waste to be descended from the surface to backfill a stope of mine m ∈ M at
time t ∈ T
13
Strategic planning of an underground mine with variable cut‑off…
percentages of dilution and recovery. Part 2 shows the total cost of both developments
and the sum of the production costs of the two mining methods. The costs of opening
and closing a mine are included in Part 3, while fixed costs to maintain a mine open are
in Part 4. Finally, the costs of additional handling for backfill material (waste material)
are represented by Part 5.
Part 1 Part 2 Part 3
⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞ ⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞ ⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞
∑ ∑ [ ∑∑ ∑ ∑
t
max Z = f Rt Nag xagt − Cag xagt − (CmI yr1m t + CmC zmt )
t∈T m∈M a∈Pm g∈G a∈Am g∈G
Part 4
⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞ Part 5
∑ t ⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞ ]
− CmM (yr1m u − zmu ) − CW (win out
mt + wmt )
u=1
(1)
3.3 Constraints
Constraints can be grouped into eight main sets. Each set is described in the following
sections.
3.3.1 Global constraints
∑∑
Nag xagt ≤ Nt t∈T
(3)
a∈P g∈G
∑ ∑
Lag xagt ≤ Lt t∈T
(4)
a∈A�H g∈G
13
R. Martinelli et al.
3.3.2 Beginning of activities
Constraints (5) limit the tasks to start at most once. They are inequalities because if
a lens is not economically exploitable, it will be left behind. However, this additional
freedom increases the difficulty of the problem. Indeed, the number of combinations is
higher when tasks can be performed or not.
∑
yat ≤ 1 a ∈ A (5)
t∈T
3.3.4 Precedence constraints
Constraints (7) and (8) represent the precedence between different activities. Con-
straints (7) ensure that a mine can only be closed if it was opened before. Con-
straints (8) ensure that each precedence constraint associated with each arc in graph
Gm = (Nm , Am ) , for each mine, is verified. It ensures that the activity associated with
the initial node is completed before the activity associated with the terminal node
begins.
t t
∑ ∑
zmu ≤ yr1,m u m ∈ M, t ∈ T (7)
u=1 u=1
t t
∑ ∑∑
ybu ≤ xagu m ∈ M, (a, b) ∈ Am , t ∈ T (8)
u=1 g∈G u=1
13
Strategic planning of an underground mine with variable cut‑off…
Constraints (9) check, for each year, that the time required to perform successive
activities does not exceed one year. For each lens 𝓁 in Fig. 2, it can be shown that
there are three sets of consecutive activities from r1,m to h2𝓁,m , which are represented
by three paths that can be distinguished by their last three activities as shown in
Table 4.
∑ ∑
Dag xagt ≤ 1 m ∈ M, 𝓁 ∈ L, p ∈ {1, 2, 3}, t ∈ T
a∈A
p
g∈G (9)
r1,m →h2
𝓁,m
3.3.6 Closure of a mine
Constraints (10) ensure that each mine closes at most once. If these constraints were
not present, the model could temporarily cease operations at a mine and return to
it later. This option is possible in practice, but it is not part of the mandate. In this
situation, other costs would be required. Indeed, temporarily closing a mine does
not eliminate all fixed costs. A cost of reopening should also be added. In this paper,
if the decision to close a mine is taken, then all related activities must stop. This
behavior is obtained using constraints (11).
∑
zmt ≤ 1 m ∈ M (10)
t∈T
t
∑ ∑
xagt ≤ 1 − zmu m ∈ M, a ∈ Am , t ∈ T (11)
g∈G u=1
There is no need to add to the MIP model constraints regarding the opening of a
mine, as all the production variables xagt are bounded to the first CAPEX develop-
ment of each mine by the constraints described in Sects. 3.3.3 and 3.3.4. Moreover,
we can always expect a mine to close as soon as its operations finish since, in prac-
tice, one year of operational costs is larger than the mine closing cost.
3.3.7 Cut‑off grades
Constraints (12) ensure that only one cut-off grade is selected per lens. Unlike pre-
vious constraints, it is an equality constraint. This restriction does not change the
solution since the lens can be assigned a cut-off without being exploited. However,
it slightly facilitates the resolution of the MIP model. Furthermore, Constraints (13)
bind activities variables with the cut-off grade. In a first version of the mathemati-
cal formulation, instead of using the v𝓁g variables, there were binary variables to
choose when to start an activity with a given cut-off grade. With the inclusion of the
v𝓁g variables and by using the binding constraints, the number of binary variables
13
R. Martinelli et al.
was reduced by about 30% for an instance of 20 lenses, seven cut-off grades, and
a period of 15 years. This proportion is larger with the increase in the number of
lenses and years. However, the number of constraints increases in the order of 10%
for an instance of the mentioned size.
∑
v𝓁g = 1 m ∈ M, 𝓁 ∈ Lm
(12)
g∈G
∑
xagt ≤ v𝓁g m ∈ M, 𝓁 ∈ Lm , a ∈ A𝓁 , g ∈ Gown (13)
t∈T
Equation (14) is a flow conservation constraint for the amount of waste material pro-
duced and used. To meet this requirement, it is necessary, for each year, that the sum
of the amount of waste rock extracted from development activities and the amount
of waste material transported from the surface is equal to the sum of the amount
of waste material used as backfill in the stopes and the amount of waste mate-
rial transported to the surface. Given that the additional quantities of waste mate-
rial descended from the surface or transported to the surface generates a cost, it is
impossible to obtain a positive value for these two quantities in the same year.
[ ] [ ]
∑ ∑ ∑∑
S in L
F Vag xagt + wmt − F Vag xagt − wout
mt
= 0 m ∈ M, t ∈ T
a∈Dm g∈G a∈Pm g∈G
(14)
4 Acceleration strategies
Despite the performance of solvers available on the market, the MIP model
described in the previous section is still challenging to solve. As a result of this, dif-
ferent strategies were devised to reduce the computation time. This section describes
some of the most successful we have found.
4.1 Preprocessing
13
Strategic planning of an underground mine with variable cut‑off…
In the procedure that calculates the earliest start, we consider the global con-
straints (2)–(4), precedence constraints (8), and the constraints on successive tasks
(9). For each activity, the procedure simulates the building of all required precedent
activities and then calculates the earliest period the activity may start. With this
information, all variables associated with the activity whose index t is lower than the
earliest moment found can be set a priori to zero. We also tested the use of the latest
start of a task as a preprocessing strategy. However, since it led to almost no impact
on the solution time, we decided not to use it.
4.2 Constructive heuristic
The constructive heuristic is a method that aims to find high-quality and feasible
solutions quickly. The solutions produced by this heuristic can then be used as hot-
start for the formulation or any other more refined heuristic. The idea to construc-
tively build a lens is that there is a natural activity order which attempts to maximize
the profit. Based on the precedence constraints, this ordering would be to start with
the Capex(es) development(s) needed to reach the lens ( r𝓁,m in Fig. 2), then to build
the first part of the Opex development ( o1𝓁,m ) in order to promptly start the profitable
cut and fill production ( c1𝓁,m ). Then, the second part of the Opex development ( o2𝓁,m )
should be performed to allow the start of the first part of the very profitable long
hole production ( h1𝓁,m ). Finally, the second part of the cut and fill production ( c2𝓁,m )
should be performed, which will allow the second part of the long hole production
( h2𝓁,m ) to start.
The heuristic creates a solution using the procedure described above over a given
ordering to the lenses. Since the procedure is quite fast, it is possible to test several
different orderings, resulting in a better feasible solution. The first strategy to obtain
an ordering considers all lenses as candidates and, at each iteration, selects the best
lens among the ones not built yet based on its estimated profit. This estimated profit
can be calculated simulating the construction of a lens for all cut-off grades over the
current partial solution as early as possible. Note that this value is not always the
same because the partial solution may differ for different given iterations, resulting
in different estimations. After the best lens is selected, it is built over the current par-
tial solution and marked as not available for future iterations.
The second strategy considers only one mine at each time. This is done because
it is very costly to maintain a mine opened, therefore in optimal solutions, usually
a mine is opened just after or just before the closing of another mine. It remains to
choose in which order to explore the mines. Since the number of mines is not large,
we decided to test all possible combinations of mines. For example, a case with five
mines, this procedure tests 5! = 120 combinations. We can reduce this number sim-
ulating the best construction of each mine independently and calculating its profit. If
the profit is not positive, the mine is not considered. Hence, for each combination, at
a given iteration, this strategy considers only the lenses of the current mine. It then
chooses the next lens to build using two different policies. The first one is the same
as used by the first strategy, choosing to build next the lens with the best estimated
profit. The other policy is to choose the lenses in the natural ordering of the mine.
13
R. Martinelli et al.
4.3 Fix‑and‑optimize heuristic
It is important to note that for every decomposition presented, the continuous var-
iables xagt , wout
mt
and win
mt
are never fixed, since they do not impose a large decline
on the solution performance. Instead, this decision improves the solution quality.
Moreover, since all the decompositions do not operate over a large part of the
solution, there is no need to limit the formulation’s execution.
13
Strategic planning of an underground mine with variable cut‑off…
The complete procedure is as follows. The procedure starts by running the Con-
structive Heuristic described in Sect. 4.2 to obtain a set of initial feasible solutions.
From this set of solutions, the Fix-and-Optimize Heuristic then performs a solution
polishing on the five best solutions to decide which one will be the initial solution.
The solution polishing is done by using the first four decompositions for all possible
periods, lenses, and mines. At the end of the solution polishing, the heuristic selects
the solution with the best net present value.
The main procedure is presented in Algorithm 1. After selecting the initial solu-
tion, the heuristic starts to explore the decompositions through a variable neigh-
borhood descent (VND) approach (Hansen and Mladenović 2001). The algorithm
uses two sets of decompositions, N1 and N2 , and at each iteration it explores all the
decompositions of N1 , but it only explores the decompositions of N2 if there was no
improvement from any (𝛿, p) ∈ N1 , where 𝛿 is the type of decomposition and p its
parameters. If all decompositions fail to improve the current solution, the procedure
ends.
It is important to note that different values for offsets are used. The decomposition
decompose-periods uses offset o1 which in the following tests is set to 4. In the
case of decomposition decompose-mines, the procedure uses offset o4 which is
set to 1. Larger values showed to be prohibitively time consuming. For the sets of
decompositions, we included all decompositions but decompose-lenses-pair
in N1 and only decompose-lenses-pair in N2.
5 Computational experiments
13
R. Martinelli et al.
The machines used to run the algorithms were AMD Opteron 6172 2.1GHz, with
32 GB of total available RAM and running Linux Operating System. The algorithms
have been developed in C++ language and compiled using the GNU g++ compiler.
For the solution of the MIP formulation, we used IBM CPLEX 12.8 solver with
default parameters. Aiming to be as close as possible to a real planning environment,
we report results for all instances using one and four threads, and considering a time
limit of 12 hours (43,200 seconds) for the MIP formulation.
We further conduct a test to check if the MIP formulation can obtain better
bounds when using the solution of the Fix-and-Optimize heuristic as a hot-start. The
best solution found by the heuristic is given to the formulation as the initial incum-
bent solution, and then the formulation is allowed to run with a time limit of 12
hours.
5.1 Comparison of methods
To assess the quality of the algorithms, the tests were performed on instances with
different quantity of mines, ranging from one to eight, 25 instances for each number,
in a total of 200 instances. Moreover, the instances contain from 4 to 75 lenses and
20 periods. The instances were generated based on real information obtained from
Glencore and performing perturbations on the data. The sizes of instances were
chosen to test the proposed methods from the smallest possible number of mines
to sizes which surpass the average number of mines found in real-life applications.
Further information about the instances are also presented in “Appendix”.
For reasons of confidentiality of company revenues, instead of presenting the
solution values, we present all results as gaps. To perform a fair comparison of solu-
tion approaches, we report two types of gaps. For all results using the MIP formula-
tion, we report the formulation gap ((best bound − integer solution)/best bound),
representing the difference between the value of the best bound (upper bound) and
the best integer solution (lower bound), both information found by the formulation.
The second gap is the gap to best, presented for all approaches, as it compares the
best integer solution found by the given approach with the best overall upper bound
found. The optimal solution for most small instances is known and can be obtained
just by running the MIP formulation, as we present in “Appendix”. In contrast, solv-
ing larger instances is a difficult task, as the MIP formulation is unable to find the
optimal solution for those instances even after a run of several hours.
Figure 3 shows the average gap (to best) for each instance size using only one
thread. MIP represents the MIP formulation results with the preprocessing strategy.
FOH and FOH+MIP represent the pure Fix-and-Optimize Heuristic and the MIP for-
mulation using the Fix-and-Optimize solution as a hot-start, respectively. From this
figure, we can notice that for a small number of mines, the MIP formulation can find
competitive results, but as this number increases, the Fix-and-Optimize Heuristic
surpasses the formulation. Moreover, we notice there is no significant improvement
when using the heuristic solution as a hot-start for the formulation, suggesting that
the integer solution obtained by the heuristic may be close to the optimal one.
13
Strategic planning of an underground mine with variable cut‑off…
Fig. 3 Average gap (to best) comparison between the solutions found by the MIP Formulation and the
Fix-and-Optimize Heuristic using one thread
Table 5 Average results for the MIP formulation and the Fix-and-Optimize Heuristic using one thread
Set MIP FOH FOH+MIP
Gown (%) Gbest (%) Twall Gbest (%) Twall Gown (%) Gbest (%) Twall
The results presented in part in Fig. 3 are summarized in Table 5. This table
shows average results for the MIP formulation (MIP), the Fix-and-Optimize Heuris-
tic (FOH), and both methods combined (FOH+MIP), using only one thread. For each
instance set and approach, columns Gown and Gbest present the average gap and aver-
age gap to best, respectively. Column Twall presents the wall time in seconds. There
is no reason to present the CPU time since it is roughly the same as the wall time
when using only one thread.
Figure 4 presents similar results, but for the experiments using four threads. The
average results obtained by the three configurations are almost the same. However,
we can notice a gap improvement found by the MIP formulation, with and without
the Fix-and-Optimize solution as a hot-start.
13
R. Martinelli et al.
Fig. 4 Average gap (to best) comparison between the solutions found by the MIP Formulation and the
Fix-and-Optimize Heuristic using four threads
Table 6 presents summary results for all configurations using four threads.
This table has almost the same arrangement of Table 5, except now we present
CPU times ( Tcpu ). Some conclusions can be drawn from these results. First,
since the improvements obtained in the MIP formulation are larger for the
gap, we may conclude the best bounds generated by the formulation are better,
a behavior that strengthens the claim that the heuristic has found high-quality
integer solutions. Second, even running the MIP formulation for 12 hours and
using 4 threads, the Fix-and-Optimize heuristic still obtains better solutions, and
in a fraction of the time (2% on average). Finally, the gain of multi-thread is not
evident since the heuristic runs only a bit faster, and there is not a large improve-
ment in the solutions obtained by the MIP formulation.
Table 6 Average results for the MIP formulation and the Fix-and-Optimize Heuristic using four threads
Set MIP FOH FOH+MIP
Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%)
1m 0.0 0.0 42.6 132.5 0.0 37.5 81.7 0.0 0.0 76.3 208.1
2m 0.0 0.0 2785.6 10707.0 0.0 114.2 256.9 0.0 0.0 3275.9 12562.0
3m 0.7 0.4 15295.2 59348.0 0.6 195.6 385.2 0.5 0.4 14807.9 57424.7
4m 4.4 3.7 34586.6 134977.1 3.7 339.9 681.0 4.0 3.6 33753.9 130573.4
5m 9.5 8.3 37903.2 147041.6 8.0 617.1 1279.0 8.2 7.9 37110.7 141826.7
6m 12.3 11.4 43203.5 165962.7 10.8 881.8 1909.0 11.3 10.6 44058.3 168549.8
7m 13.3 12.1 43203.6 165306.7 11.8 1042.9 2106.6 12.2 11.7 44211.2 167676.8
8m 19.3 17.8 43204.1 164025.5 16.6 1412.7 2601.0 16.7 16.3 44630.9 167438.0
Avg 7.4 6.7 27528.1 105937.6 6.5 580.2 1162.5 6.6 6.3 27740.6 105782.4
13
(a) (b)
Strategic planning of an underground mine with variable cut‑off…
(c) (d)
13
Fig. 5 Evolution of best integer solution and dual bound for chosen instances
R. Martinelli et al.
5.2 Sensitivity analysis
Based on the results presented in the last section, the algorithm chosen to perform
the sensitivity analysis is the Fix-and-Optimize heuristic (FOH), for it obtained the
best solutions in the smallest amount of time. The solutions obtained by this con-
figuration are considered as 100%.
Fixed cut-off grades The first analysis is to prove the impact of one of the con-
tributions of our work. As previously stated, in real-life, the selection of the cut-off
grade is usually made before planning. This strategy is expected to lead to subopti-
mal solutions. For this reason, we test all instances with a fixed cut-off grade value
out of the seven possible choices. Figure 6 presents the results for this test.
The best overall fixed cut-off grade was COG5, yielding an average solution qual-
ity of 82.4%, with worse values for all other choices. This result indicates that using
Fig. 6 Solution quality comparison between the variable cut-off grade and when it is fixed to a given
value
13
Strategic planning of an underground mine with variable cut‑off…
Time Quality Time Quality Time Quality Time Quality Time Quality Time
(%) (%) (%) (%)
1m 50.8 91.87 9.2 91.57 13.8 79.68 0.6 99.99 29.1 91.49 36.4
2m 168.7 99.05 37.5 94.69 52.2 89.43 2.3 99.98 82.6 94.59 91.8
3m 282.1 98.89 69.2 97.62 93.5 94.50 3.7 100.03 129.4 98.47 163.9
4m 465.5 99.08 104.7 98.78 164.9 94.83 5.3 99.76 177.4 99.05 328.6
5m 803.3 98.82 141.0 99.13 272.5 94.41 8.1 99.78 339.7 99.29 624.1
6m 1251.8 98.69 204.2 99.09 415.7 93.23 14.7 99.56 448.4 98.53 888.0
7m 1284.4 98.55 228.1 99.00 476.5 93.69 31.7 99.40 381.7 98.84 922.5
8m 1709.5 97.63 459.9 97.26 764.2 90.30 209.7 98.61 618.3 97.69 1448.6
Avg 752.0 97.82 156.7 97.14 281.7 91.26 34.5 99.64 275.8 97.24 563.0
Table 8 Solution quality change when using all but one Fix-and-Optimize decomposition
Set FOH Periods Lens Cutoff-grades Mine Lenses-pair
Time Quality Time Quality Time Quality Time Quality Time Quality Time
(%) (%) (%) (%) (%)
1m 50.8 100.00 51.6 100.00 46.7 100.00 53.4 96.12 35.5 100.00 37.5
2m 168.7 100.00 161.9 99.88 141.3 100.00 179.3 99.74 116.8 100.00 140.3
3m 282.1 100.01 253.7 99.99 239.3 100.00 291.1 99.23 217.8 99.97 204.7
4m 465.5 99.79 500.9 100.06 389.9 100.00 489.2 99.49 326.8 99.95 310.6
5m 803.3 100.02 936.0 100.05 765.1 100.00 824.7 99.82 610.8 99.98 577.0
6m 1251.8 99.78 1271.3 99.86 947.8 99.94 1125.4 99.66 881.7 99.94 810.4
7m 1284.4 99.91 1235.7 99.83 964.1 100.00 1244.1 99.88 1039.5 99.97 807.3
8m 1709.5 99.61 1638.8 99.44 1338.5 99.68 1583.8 99.44 1460.1 99.94 1204.0
Avg 752.0 99.89 756.2 99.89 604.1 99.95 723.9 99.17 586.1 99.97 511.5
a fixed cut-off grade for all lenses may incur a substantial revenue loss. Consider-
ing the dimension of mining operations, the variable cut-off grade optimization may
lead to huge savings.
Fix-and-optimize decompositions The second sensitivity analysis is to show
the importance of each decomposition type. First, we tested all instances using
only one decomposition. The results obtained are presented in Table 7, where
for each instance set, column FOH Time shows the times for the complete tests.
The following pairs of columns show the average solution quality and the aver-
age time when running the Fix-and-Optimize heuristic with only one decompo-
sition. From this table, it is possible to conclude that decomposition cutoff-
grades leads to the worst average solution quality, but in a small amount of
time. Furthermore, the best decomposition is mine, which can obtain the best
overall solution quality. Nevertheless, these results show that all the decomposi-
tions are useful.
13
R. Martinelli et al.
13
Strategic planning of an underground mine with variable cut‑off…
Table 9 (continued)
Instance Number of lenses MIP formulation
Total In mine Variables Constraints Non-zeros
13
R. Martinelli et al.
Table 9 (continued)
Instance Number of lenses MIP formulation
Total In mine Variables Constraints Non-zeros
13
Strategic planning of an underground mine with variable cut‑off…
Table 9 (continued)
Instance Number of lenses MIP formulation
Total In mine Variables Constraints Non-zeros
13
R. Martinelli et al.
Table 9 (continued)
Instance Number of lenses MIP formulation
Total In mine Variables Constraints Non-zeros
13
Strategic planning of an underground mine with variable cut‑off…
Table 9 (continued)
Instance Number of lenses MIP formulation
Total In mine Variables Constraints Non-zeros
6 Conclusion
In this paper, a MIP model for the long term planning of an underground mine with a
variable cut-off grade was presented. It has been shown that the resolution of this model
is prohibitively expensive when using a commercial library for MIP models in real-life
cases. For this reason, heuristic approaches were proposed to attempt to reduce solution
times. The experimental results showed that the heuristics could provide high-quality
13
R. Martinelli et al.
solutions fast. The presented results confirmed that significant improvements for the
solution of the problem could be obtained by using the proposed methods. Moreover,
the results also show the methods to be scalable for instances with the same character-
istics of Raglan mine, since real-life applications are smaller than the largest instances
tested. Finally, letting the optimizer determine the best cut-off grade to use for each lens
has increased the net present value by more than 20% on average.
From the real-life perspective, long-term planning for underground mines is
an annual exercise conducted by mining companies due to, among other things,
uncertainties related to deposit geology and fluctuating metal prices. It is gener-
ally done by studying only two or three scenarios. The study of these scenarios can
take 1 to 2 months of work for a team of engineers. This study takes into account
all the engineering aspects of mine operations. The long-term planning tool that
we propose allows us to evaluate all the feasible scenarios and to select the best
mining sequence. Despite some slight simplifications, the tool developed can be
handy by the engineers of the mine since they can start from this optimal scenario
to complete the study with more details, which saves much time. On the other
hand, if the optimal scenario is not appropriate, it will be easy to add additional
constraints to the model to converge to a more realistic and optimal scenario.
Furthermore, the approaches are flexible and fast, allowing mining engineers
to test several different scenarios. These scenarios can be built from different lev-
els of cut-off grade discretization or just one fixed cut-off grade. They can also
be built using other mining methods but the ones considered in the presented
application, using different precedence rules and split strategies. The activities
can be split in any number of parts, with uneven attributes (i.e., a split in two
does not need to be 50%/50%). Finally, the planning may be revised whenever
needed, considering the current state of each mine and available equipment.
Acknowledgements This research was partially supported by the Fonds de recherche du Québec - Nature
et technologies (FRQ-NT) and by the Conselho Nacional de Desenvolvimento Científico e Tecnológico
(CNPq), grants 313521/2017-4 and 425962/2016-4. All support is gratefully acknowledged. Computa-
tions were made on the supercomputer Mp2, managed by Calcul Québec and Compute Canada. The oper-
ation of this supercomputer is funded by the Canada Foundation for Innovation (CFI), the ministère de
l’Économie, de la science et de l’innovation du Québec (MESI) and the Fonds de recherche du Québec
- Nature et technologies (FRQ-NT). The authors would also like to thank the anonymous reviewers for
their helpful and constructive comments that contributed to improve the final version of this paper, and
Thibaut Vidal for helping with the figures.
In this section, we present detailed information and results for all mine sizes and all
approaches presented throughout the paper. Table 9 presents the number of lenses for
each instance, in total and for each mine, and the number of variables, constraints,
and non-zeros present in the MIP formulation. In Tables 10, 11, 12, 13, 14, 15, 16
and 17, we present the results for the original MIP formulation with the preprocess-
ing procedure, for the Fix-and-Optimize Heuristic and for the MIP formulation using
the Fix-and-Optimize heuristic solution as a hot-start. All results are presented using
one thread and four threads, with a time limit of 12 hours for the MIP formulation.
13
Table 10 Results for 1-mine instances
Inst One thread Four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
1m-01 0.0 0.0 10.3 0.0 16.5 0.0 0.0 25.8 0.0 0.0 6.1 19.4 0.0 11.1 16.7 0.0 0.0 18.2 38.8
1m-02 0.0 0.0 8.5 0.0 14.5 0.0 0.0 21.4 0.0 0.0 9.1 31.8 0.0 10.1 15.6 0.0 0.0 15.2 30.9
1m-03 0.0 0.0 23.2 0.0 52.1 0.0 0.0 66.8 0.0 0.0 10.8 33.3 0.0 37.0 98.3 0.0 0.0 40.9 119.1
1m-04 0.0 0.0 75.6 0.0 32.4 0.0 0.0 81.6 0.0 0.0 24.9 76.3 0.0 28.0 62.0 0.0 0.0 48.3 114.9
1m-05 0.0 0.0 30.4 0.0 30.0 0.0 0.0 59.1 0.0 0.0 14.0 33.7 0.0 26.7 47.8 0.0 0.0 41.3 83.9
1m-06 0.0 0.0 29.7 0.0 16.5 0.0 0.0 52.2 0.0 0.0 18.6 69.2 0.0 13.7 21.2 0.0 0.0 24.3 58.8
1m-07 0.0 0.0 16.0 0.0 17.5 0.0 0.0 32.3 0.0 0.0 14.0 36.3 0.0 13.8 21.1 0.0 0.0 21.6 37.5
1m-08 0.0 0.0 28.8 0.0 32.5 0.0 0.0 64.7 0.0 0.0 12.2 32.8 0.0 23.8 42.8 0.0 0.0 34.5 68.7
1m-09 0.0 0.0 75.0 0.0 31.5 0.0 0.0 92.7 0.0 0.0 32.5 94.5 0.0 26.6 50.2 0.0 0.0 56.1 131.5
1m-10 0.0 0.0 35.4 0.0 32.4 0.0 0.0 61.4 0.0 0.0 22.4 64.2 0.0 56.0 100.8 0.0 0.0 40.9 96.0
1m-11 0.0 0.0 50.3 0.0 27.4 0.0 0.0 74.7 0.0 0.0 27.2 79.7 0.0 21.2 37.9 0.0 0.0 44.8 101.5
Strategic planning of an underground mine with variable cut‑off…
1m-12 0.0 0.0 155.5 0.0 54.9 0.0 0.0 265.2 0.0 0.0 47.9 168.1 0.0 36.7 84.2 0.0 0.0 92.1 296.9
1m-13 0.0 0.0 49.0 0.0 33.6 0.0 0.0 74.8 0.0 0.0 21.1 66.3 0.0 53.5 95.2 0.0 0.0 42.4 96.2
1m-14 0.0 0.0 38.2 0.0 66.5 0.0 0.0 106.7 0.0 0.0 19.7 64.5 0.0 39.3 88.4 0.0 0.0 55.9 142.4
1m-15 0.0 0.0 21.3 0.0 32.0 0.0 0.0 49.5 0.0 0.0 13.9 34.5 0.0 24.5 46.2 0.0 0.0 38.3 76.6
1m-16 0.0 0.0 114.9 0.0 111.1 0.0 0.0 202.5 0.0 0.0 41.3 131.7 0.0 58.1 137.3 0.0 0.0 96.4 253.0
1m-17 0.0 0.0 101.3 0.0 77.2 0.0 0.0 196.0 0.0 0.0 35.9 111.9 0.0 55.9 124.4 0.0 0.0 94.3 245.1
1m-18 0.0 0.0 43.5 0.0 38.9 0.0 0.0 72.7 0.0 0.0 22.3 58.5 0.0 37.6 76.5 0.0 0.0 57.3 129.6
1m-19 0.0 0.0 97.7 0.0 52.2 0.0 0.0 203.8 0.0 0.0 44.6 136.7 0.0 35.6 84.8 0.0 0.0 71.4 190.3
1m-20 0.0 0.0 246.8 0.0 145.3 0.0 0.0 377.1 0.0 0.0 189.8 633.8 0.0 74.0 179.9 0.0 0.0 223.5 683.4
1m-21 0.0 0.0 89.3 0.0 102.3 0.0 0.0 217.6 0.0 0.0 50.4 167.7 0.0 58.2 132.6 0.0 0.0 118.4 333.3
1m-22 0.0 0.0 86.2 0.0 41.3 0.0 0.0 116.3 0.0 0.0 36.3 111.6 0.0 36.7 74.6 0.0 0.0 80.4 211.5
1m-23 0.0 0.0 39.3 0.0 26.4 0.0 0.0 68.1 0.0 0.0 17.7 40.4 0.0 26.2 43.3 0.0 0.0 44.6 87.2
13
Table 10 (continued)
Inst One thread Four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
1m-24 0.0 0.0 147.2 0.0 40.4 0.0 0.0 181.8 0.0 0.0 67.7 205.4 0.0 29.6 59.3 0.0 0.0 108.1 305.4
1m-25 0.0 0.0 640.6 0.0 144.1 0.0 0.0 673.1 0.0 0.0 265.5 808.8 0.0 103.4 302.1 0.0 0.0 399.4 1269.6
Avg 0.0 0.0 90.2 0.0 50.8 0.0 0.0 137.5 0.0 0.0 42.6 132.5 0.0 37.5 81.7 0.0 0.0 76.3 208.1
R. Martinelli et al.
Table 11 Results for 2-mine instances
Inst One thread Four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
2m-01 0.0 0.0 24.7 0.0 32.9 0.0 0.0 52.0 0.0 0.0 21.5 52.4 0.0 27.9 37.1 0.0 0.0 47.0 84.2
2m-02 0.0 0.0 397.6 0.0 81.1 0.0 0.0 359.2 0.0 0.0 113.0 387.4 0.0 51.1 105.3 0.0 0.0 140.2 414.5
2m-03 0.0 0.0 420.4 0.0 38.6 0.0 0.0 297.2 0.0 0.0 112.4 392.3 0.0 25.1 43.0 0.0 0.0 124.7 390.4
2m-04 0.0 0.0 1256.3 0.0 78.8 0.0 0.0 1666.6 0.0 0.0 833.3 2898.0 0.0 54.7 95.8 0.0 0.0 776.3 2681.2
2m-05 0.0 0.0 21.7 0.0 59.4 0.0 0.0 76.5 0.0 0.0 22.5 57.9 0.0 46.4 70.1 0.0 0.0 63.2 110.4
2m-06 0.0 0.0 906.5 0.0 65.5 0.0 0.0 773.0 0.0 0.0 566.4 1974.0 0.0 51.8 94.3 0.0 0.0 753.5 2612.4
2m-07 0.0 0.0 46.7 0.0 74.0 0.0 0.0 117.0 0.0 0.0 25.8 78.3 0.0 61.6 116.4 0.0 0.0 86.8 189.4
2m-08 0.0 0.0 61.0 0.0 64.2 0.0 0.0 119.1 0.0 0.0 29.9 75.9 0.0 56.4 75.1 0.0 0.0 81.2 141.8
2m-09 0.0 0.0 2534.8 0.0 350.0 0.0 0.0 3414.3 0.0 0.0 2158.2 8350.5 0.0 175.9 471.6 0.0 0.0 1472.2 5434.2
2m-10 0.0 0.0 333.4 0.0 148.4 0.0 0.0 625.0 0.0 0.0 131.7 475.0 0.0 91.6 178.4 0.0 0.0 197.6 573.4
2m-11 0.0 0.0 3044.0 0.0 123.7 0.0 0.0 2640.5 0.0 0.0 1117.3 3748.7 0.0 98.6 199.2 0.0 0.0 1338.2 4635.7
Strategic planning of an underground mine with variable cut‑off…
2m-12 0.0 0.0 1367.5 0.0 78.6 0.0 0.0 2030.2 0.0 0.0 582.7 2030.3 0.0 69.3 135.8 0.0 0.0 1568.2 5719.6
2m-13 0.0 0.0 2209.8 0.0 116.1 0.0 0.0 2644.0 0.0 0.0 1381.1 5152.4 0.0 87.1 156.7 0.0 0.0 2204.3 8106.4
2m-14 0.0 0.0 8017.5 0.0 148.1 0.0 0.0 5380.2 0.0 0.0 3203.3 11871.5 0.0 88.7 188.5 0.0 0.0 2864.8 10453.2
2m-15 0.0 0.0 179.4 0.0 78.4 0.0 0.0 196.7 0.0 0.0 75.8 253.8 0.0 80.2 128.9 0.0 0.0 131.2 308.2
2m-16 0.0 0.0 497.0 0.0 148.3 0.0 0.0 505.7 0.0 0.0 142.7 462.9 0.0 117.6 207.8 0.0 0.0 235.3 596.2
2m-17 0.0 0.0 3363.0 0.0 91.8 0.0 0.0 3130.4 0.0 0.0 1094.8 3878.5 0.0 69.2 111.8 0.0 0.0 1295.9 4706.9
2m-18 0.0 0.0 8284.1 0.0 315.3 0.0 0.0 7817.0 0.0 0.0 3015.1 11624.0 0.0 253.1 685.5 0.0 0.0 2873.9 10866.2
2m-19 0.0 0.0 1831.3 0.0 232.1 0.0 0.0 2136.0 0.0 0.0 884.3 3097.0 0.0 125.5 278.9 0.0 0.0 1200.4 4167.3
2m-20 0.0 0.0 2179.9 0.0 397.7 0.0 0.0 2292.2 0.0 0.0 296.2 1079.5 0.0 207.1 482.0 0.0 0.0 484.5 1505.4
2m-21 0.2 0.0 43200.1 0.0 417.6 0.2 0.0 43676.6 0.0 0.0 8827.2 34235.6 0.0 330.0 916.3 0.0 0.0 12404.5 48563.2
2m-22 0.0 0.0 6180.9 1.0 153.6 0.0 0.0 4815.1 0.0 0.0 24723.5 96116.3 1.0 126.8 272.4 0.0 0.0 4109.9 15720.0
13
Table 11 (continued)
Inst One thread Four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
2m-23 0.0 0.0 21808.8 0.0 290.6 0.0 0.0 12567.8 0.0 0.0 7530.0 29373.1 0.0 187.8 462.8 0.0 0.0 7230.4 27838.4
2m-24 0.6 0.0 43200.1 0.0 530.6 0.0 0.0 34710.3 0.0 0.0 12037.8 47627.9 0.0 296.4 769.4 0.0 0.0 39066.3 154396.4
2m-25 0.0 0.0 1678.0 0.0 102.8 0.0 0.0 1933.8 0.0 0.0 714.0 2382.8 0.0 75.4 139.2 0.0 0.0 1146.6 3833.8
Avg 0.0 0.0 6121.8 0.0 168.7 0.0 0.0 5359.1 0.0 0.0 2785.6 10707.0 0.0 114.2 256.9 0.0 0.0 3275.9 12562.0
R. Martinelli et al.
Table 12 Results for 3-mine instances
Inst One thread Four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
3m-01 0.0 0.0 190.0 0.0 89.0 0.0 0.0 263.8 0.0 0.0 92.0 323.3 0.0 63.4 105.8 0.0 0.0 128.8 328.9
3m-02 0.0 0.0 1019.1 0.0 78.8 0.0 0.0 1809.9 0.0 0.0 482.5 1771.1 0.0 63.6 108.7 0.0 0.0 235.2 741.9
3m-03 0.0 0.0 301.7 0.0 83.1 0.0 0.0 297.9 0.0 0.0 130.6 448.8 0.0 71.0 120.9 0.0 0.0 155.7 399.2
3m-04 0.0 0.0 759.5 0.0 202.7 0.0 0.0 1584.4 0.0 0.0 710.0 2531.2 0.0 124.3 271.5 0.0 0.0 630.7 2038.3
3m-05 0.0 0.0 681.9 0.0 116.7 0.0 0.0 743.9 0.0 0.0 340.5 1077.1 0.0 107.2 165.5 0.0 0.0 559.9 1407.5
3m-06 0.0 0.0 795.8 0.0 105.8 0.0 0.0 1030.3 0.0 0.0 298.1 1034.7 0.0 97.5 162.1 0.0 0.0 452.0 1410.2
3m-07 1.3 0.0 43200.1 0.1 226.8 0.0 0.0 16307.1 0.0 0.0 15602.4 60119.2 0.1 177.4 369.6 0.0 0.0 6314.0 24247.5
3m-08 0.0 0.0 1636.2 0.0 144.2 0.0 0.0 2170.6 0.0 0.0 319.9 1179.7 0.0 105.9 186.8 0.0 0.0 358.7 1125.7
3m-09 0.0 0.0 27081.2 0.0 442.4 0.0 0.0 20457.9 0.0 0.0 20373.9 77692.5 0.0 245.1 622.4 0.0 0.0 24674.8 96458.5
3m-10 1.8 1.0 43200.1 1.0 260.1 1.0 1.0 43469.2 1.7 1.0 43204.7 168997.3 1.1 196.9 437.1 1.5 1.0 43404.5 169481.5
3m-11 0.0 0.0 3407.3 0.0 345.5 0.0 0.0 5992.5 0.0 0.0 1757.8 5881.7 0.0 256.5 479.2 0.0 0.0 2605.3 8793.4
Strategic planning of an underground mine with variable cut‑off…
3m-12 0.1 0.0 43200.1 0.6 293.3 1.1 0.0 43477.9 0.0 0.0 24852.1 96663.1 0.6 207.8 435.4 0.1 0.0 43408.3 170437.0
3m-13 0.0 0.0 8757.0 0.0 174.7 0.0 0.0 5906.7 0.0 0.0 2154.8 8095.1 0.0 123.6 235.7 0.0 0.0 1995.7 7307.6
3m-14 5.0 2.3 43200.1 2.9 253.8 4.5 1.0 43467.3 1.5 1.0 43200.9 170509.6 2.1 231.5 488.3 1.0 1.0 43429.6 170896.7
3m-15 0.0 0.0 866.4 0.0 195.1 0.0 0.0 895.2 0.0 0.0 792.4 2636.8 0.0 176.0 302.1 0.0 0.0 548.4 1682.7
3m-16 4.1 3.5 43200.1 3.4 447.8 5.3 3.4 43633.5 3.4 3.4 43200.2 170682.2 3.4 307.3 566.1 3.8 3.4 43500.1 170812.5
3m-17 0.0 0.0 416.6 0.0 278.2 0.0 0.0 592.2 0.0 0.0 135.9 467.3 0.0 184.5 302.3 0.0 0.0 290.9 668.1
3m-18 0.0 0.0 509.3 0.0 140.8 0.0 0.0 1070.1 0.0 0.0 399.2 1198.8 0.0 112.3 163.3 0.0 0.0 494.4 1377.5
3m-19 1.5 1.3 43200.1 1.5 829.0 3.3 1.5 43968.6 2.1 1.3 43200.2 170839.2 1.5 316.7 771.5 1.3 1.3 43518.6 169749.4
3m-20 0.0 0.0 707.6 0.0 345.8 0.0 0.0 1004.2 0.0 0.0 722.1 2057.8 0.0 199.8 357.1 0.0 0.0 680.9 1914.2
3m-21 0.0 0.0 11763.0 2.3 284.4 0.4 0.0 43466.0 0.3 0.0 43209.5 159024.7 2.3 214.8 470.3 0.0 0.0 6042.3 23115.0
3m-22 9.2 4.2 43200.2 4.2 530.4 8.1 4.2 43755.6 7.9 4.4 43204.9 170301.6 4.2 329.5 671.0 4.2 4.2 43539.2 169147.0
3m-23 1.9 0.0 43200.1 0.2 412.8 2.0 0.2 43595.2 0.0 0.0 35714.4 139585.7 0.2 266.1 494.2 0.2 0.0 43471.6 169151.0
13
Table 12 (continued)
Inst One thread Four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
3m-24 0.0 0.0 17121.0 0.0 511.4 0.0 0.0 17053.5 0.0 0.0 14045.3 54553.5 0.0 494.8 977.9 0.0 0.0 12936.0 48419.3
3m-25 0.0 0.0 9122.1 0.0 259.3 0.0 0.0 10607.0 0.0 0.0 4235.7 16029.3 0.0 215.7 365.1 0.0 0.0 6820.7 24506.6
Avg 1.0 0.5 17229.5 0.7 282.1 1.0 0.5 17464.8 0.7 0.4 15295.2 59348.0 0.6 195.6 385.2 0.5 0.4 14807.9 57424.7
R. Martinelli et al.
Table 13 Results for 4-mine instances
Inst One thread Four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
4m-01 0.0 0.0 3073.2 0.0 108.8 0.0 0.0 4953.2 0.0 0.0 1325.9 4986.3 0.0 154.1 246.5 0.0 0.0 1594.0 5895.7
4m-02 0.0 0.0 14796.3 0.0 266.5 0.0 0.0 12726.8 0.0 0.0 19873.5 78528.3 0.0 208.7 350.0 0.0 0.0 7704.7 29702.6
4m-03 0.0 0.0 34105.7 0.1 201.8 3.5 0.1 43395.9 0.0 0.0 8939.3 34691.3 0.1 254.4 424.5 0.0 0.0 8713.8 33535.1
4m-04 2.8 0.4 43200.1 0.4 205.6 2.1 0.4 43408.2 0.4 0.4 43200.2 167400.2 0.4 136.9 262.7 0.5 0.4 43334.1 171068.5
4m-05 4.0 0.0 43200.1 0.0 417.8 0.9 0.0 43602.2 0.0 0.0 42711.7 165085.4 0.0 244.8 544.3 0.1 0.0 43443.4 172258.2
4m-06 7.5 5.8 43200.1 5.8 385.4 5.8 5.8 43588.7 7.0 5.8 43200.2 169732.6 5.8 281.0 595.5 6.5 5.8 43471.9 169858.5
4m-07 4.5 4.5 43200.2 4.7 240.5 5.5 4.6 43454.8 6.0 4.5 43215.8 169604.3 4.7 198.1 341.5 4.8 4.5 43529.6 156373.6
4m-08 5.2 4.9 43200.1 5.9 282.9 6.0 4.9 43507.2 4.9 4.9 43216.8 170857.0 5.9 230.4 478.6 5.8 4.9 43425.4 168700.0
4m-09 3.6 1.9 43200.1 2.0 379.5 2.5 1.9 43588.7 2.2 1.9 43203.7 170469.0 2.0 281.9 565.9 1.9 1.9 43501.5 169861.1
4m-10 0.0 0.0 34887.6 0.0 303.4 0.0 0.0 33942.4 0.0 0.0 12126.4 47286.7 0.0 206.7 424.9 0.0 0.0 15727.3 61256.5
4m-11 0.0 0.0 2030.8 0.0 195.9 0.0 0.0 1835.5 0.0 0.0 994.6 3559.4 0.0 183.8 270.4 0.0 0.0 1274.5 4210.4
Strategic planning of an underground mine with variable cut‑off…
4m-12 12.3 5.0 43200.1 5.1 444.6 11.3 5.1 43641.3 5.1 4.9 43206.0 170120.5 5.1 280.5 663.2 5.0 5.0 43484.3 170745.4
4m-13 8.2 5.6 43200.2 6.3 697.2 10.5 6.3 43880.6 5.3 5.3 43200.2 167048.5 6.3 371.0 849.0 10.7 6.3 43531.3 170410.7
4m-14 3.8 0.0 43200.1 0.0 297.5 1.1 0.0 43494.4 3.4 0.0 43200.2 168794.3 0.0 229.4 418.3 0.0 0.0 23053.3 90656.4
4m-15 0.0 0.0 1501.9 0.0 209.2 0.0 0.0 1731.7 0.0 0.0 1047.5 3868.1 0.0 187.5 334.8 0.0 0.0 1097.9 3764.4
4m-16 8.9 8.9 43200.2 8.2 468.5 9.4 8.2 43594.2 9.6 8.6 43200.3 168515.1 8.2 324.3 649.0 8.4 7.6 43525.5 170725.2
4m-17 9.1 9.1 43200.2 9.6 530.8 9.7 8.6 43668.3 8.6 8.6 43200.2 169470.3 9.6 346.9 661.1 11.2 9.6 43544.3 169177.7
4m-18 5.7 4.9 43200.2 4.8 697.2 5.8 4.8 43879.1 6.8 5.8 43200.3 169553.4 4.9 386.1 836.8 4.9 4.9 43577.5 171243.9
4m-19 15.0 11.2 43200.2 8.6 919.5 12.4 8.6 44065.2 15.2 11.0 43200.3 167693.3 8.6 678.6 1451.1 8.6 8.6 43869.1 164351.9
4m-20 1.8 1.3 43200.2 1.3 716.0 1.6 1.3 43891.4 1.5 1.3 43200.3 169864.2 1.3 685.0 1323.1 1.3 1.3 43867.9 171622.3
4m-21 18.0 13.8 43200.1 11.9 533.5 16.2 11.9 43718.0 13.6 12.4 43200.3 167666.2 11.9 422.5 848.5 11.9 11.9 43619.3 168732.6
4m-22 9.5 7.2 43200.2 7.1 910.8 10.3 7.1 44159.3 6.9 6.9 43200.4 169180.8 7.1 700.3 1452.9 8.5 7.1 43900.4 169493.8
4m-23 5.4 3.8 43200.2 4.4 541.9 5.7 3.6 43785.5 5.2 3.5 43200.4 169578.1 4.4 460.2 828.6 3.6 3.6 43646.8 165681.2
13
Table 13 (continued)
Inst One thread Four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
4m-24 6.4 3.1 43200.2 3.1 992.6 6.1 2.9 44200.7 3.1 2.8 43200.4 162449.8 3.2 536.5 1237.8 2.6 2.6 43741.2 165931.0
4m-25 6.3 4.2 43200.2 3.9 689.3 4.8 3.9 43932.5 5.1 3.9 43200.4 168424.2 3.9 507.4 966.0 3.9 3.9 43669.7 169077.6
Avg 5.5 3.8 36447.9 3.7 465.5 5.2 3.6 37185.8 4.4 3.7 34586.6 134977.1 3.7 339.9 681.0 4.0 3.6 33753.9 130573.4
R. Martinelli et al.
Table 14 Results for 5-mine instances
Inst One thread Four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
5m-01 6.7 0.0 43200.1 0.0 170.5 0.0 0.0 31668.3 0.0 0.0 25475.6 99768.3 0.0 134.3 225.6 0.0 0.0 20654.0 81469.5
5m-02 0.0 0.0 2192.3 0.1 140.2 0.0 0.0 3397.8 0.0 0.0 1297.2 4835.3 0.1 122.1 194.7 0.0 0.0 1366.3 4764.6
5m-03 0.0 0.0 12737.7 0.0 108.5 0.0 0.0 31239.7 0.0 0.0 6066.0 23051.3 0.0 96.2 155.4 0.0 0.0 6419.9 24564.7
5m-04 7.0 6.5 43200.1 5.8 236.0 7.4 5.8 43460.2 5.8 5.8 43205.1 170324.7 5.8 202.6 355.2 5.8 5.8 43406.2 170398.0
5m-05 0.0 0.0 24556.6 0.0 396.0 0.0 0.0 14886.8 1.5 0.0 43210.2 169768.2 0.0 373.7 688.6 0.0 0.0 17565.3 68461.4
5m-06 0.0 0.0 7018.1 0.4 225.8 0.0 0.0 9529.5 0.0 0.0 7485.8 27324.3 0.4 198.4 340.6 0.0 0.0 3938.2 14516.1
5m-07 9.1 5.4 43200.2 5.3 287.3 6.9 5.3 43525.3 5.3 5.3 43205.7 170013.8 5.3 219.9 407.8 6.7 5.3 43560.9 158545.4
5m-08 10.4 10.1 43200.2 9.8 534.7 10.0 9.8 43736.4 11.3 9.5 43204.1 169698.4 9.8 347.4 706.3 9.8 9.8 43560.4 168651.1
5m-09 7.1 4.1 43200.2 3.5 404.3 6.9 3.5 43622.0 6.0 3.5 43200.2 165167.3 3.5 258.6 432.2 3.4 3.4 43463.3 167266.3
5m-10 5.1 4.5 43200.2 4.4 447.8 4.5 4.4 43625.1 6.1 4.8 43204.5 169409.7 4.4 395.9 680.8 4.4 4.4 43581.7 171219.8
5m-11 12.5 7.2 43200.2 8.0 509.6 14.1 8.0 43694.1 15.7 7.2 43200.2 169287.0 8.0 374.2 744.7 7.2 7.2 43588.1 170573.6
Strategic planning of an underground mine with variable cut‑off…
5m-12 10.8 10.5 43200.2 10.3 675.4 10.2 10.2 43936.3 12.3 11.5 43203.7 167799.5 10.3 501.7 1040.8 12.2 10.2 43694.7 169413.7
5m-13 12.8 10.6 43200.2 10.3 969.5 12.0 10.3 44224.7 12.0 10.9 43200.3 165629.7 10.3 989.3 2182.4 10.3 10.3 44180.6 161871.6
5m-14 11.3 10.0 43200.3 10.8 1393.9 10.6 10.6 44661.4 10.4 10.3 43200.8 167067.2 10.8 978.3 1785.2 11.8 10.6 43786.4 169790.1
5m-15 15.0 12.2 43200.3 10.9 1119.7 13.5 10.9 44374.5 13.1 10.3 43200.4 168995.3 10.9 957.0 2125.7 10.9 10.9 44282.9 165510.5
5m-16 16.9 16.6 43200.2 12.4 1070.3 12.8 12.4 44238.4 14.6 13.9 43200.4 166255.4 12.7 655.6 1639.0 12.7 12.7 43855.4 167306.3
5m-17 14.4 2.9 43200.2 4.8 706.7 8.1 2.9 44027.7 4.6 2.9 43204.5 168218.7 4.8 579.4 1013.6 2.9 2.9 43768.8 168963.3
5m-18 17.1 16.0 43200.2 14.8 991.8 16.7 14.8 44126.3 15.8 15.4 43200.4 167145.9 14.8 583.3 1097.2 14.8 14.8 43786.7 163724.5
5m-19 21.1 20.6 43200.3 15.2 1559.2 15.2 15.2 44590.5 17.7 16.9 43200.4 164263.4 15.2 955.1 2133.1 16.4 15.2 44154.9 165338.3
5m-20 17.9 16.9 43200.3 13.1 1377.0 13.0 13.0 44586.9 14.3 13.1 43200.3 164806.8 13.1 1170.4 2820.8 14.0 13.1 44344.6 169456.6
5m-21 19.0 17.6 43200.2 13.3 1520.3 13.6 13.3 44692.4 15.2 15.2 43200.4 166653.5 13.3 891.4 2082.3 14.1 13.3 44166.3 169823.1
5m-22 21.4 18.9 43200.2 17.5 1893.9 18.9 17.5 45048.9 22.2 19.6 43207.4 169097.7 17.5 1392.5 3168.3 17.5 17.5 44539.2 164130.0
5m-23 14.4 13.8 43200.3 13.3 799.2 13.3 13.3 43986.0 16.4 15.0 43205.1 168385.8 13.3 672.6 1438.3 13.5 13.2 43883.0 170734.8
13
Table 14 (continued)
Inst One thread Four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
5m-24 10.7 10.5 43200.3 10.1 1331.7 10.1 10.1 44495.2 11.9 11.1 43200.3 165431.5 10.1 972.6 2061.2 10.2 10.1 44197.0 169185.8
5m-25 6.5 5.7 43200.3 5.6 1213.3 5.7 5.6 44328.7 6.2 5.8 43200.4 167641.6 5.6 1404.2 2455.3 5.6 5.6 44022.5 169988.9
Avg 10.7 8.8 38148.4 8.0 803.3 8.9 7.9 38948.1 9.5 8.3 37903.2 147041.6 8.0 617.1 1279.0 8.2 7.9 37110.7 141826.7
R. Martinelli et al.
Table 15 Results for 6-mine instances
Inst One thread Four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
6m-01 4.3 3.9 43200.2 3.7 344.4 5.6 3.7 43597.7 4.0 3.5 43208.2 168703.4 3.5 293.4 537.2 3.5 3.5 43499.3 170191.6
6m-02 3.4 1.8 43200.1 3.0 360.8 6.6 1.8 43533.2 1.8 1.8 43209.0 171017.4 1.8 302.3 503.4 2.4 1.8 43509.6 170701.2
6m-03 9.3 7.6 43200.1 7.4 594.8 9.6 7.4 43802.7 8.1 7.9 43204.2 166935.1 7.4 402.5 753.5 7.4 7.4 43890.2 162407.5
6m-04 9.8 5.7 43200.2 5.2 471.2 8.4 5.2 43653.3 7.2 5.8 43205.0 169192.8 5.2 339.3 608.9 5.2 5.2 43539.3 169343.1
6m-05 3.6 0.2 43200.2 0.2 356.8 2.7 0.2 43584.8 0.2 0.2 43200.2 168760.9 0.2 316.8 493.9 0.8 0.2 43517.3 170250.5
6m-06 10.6 7.5 43200.2 5.5 570.0 8.3 5.0 43758.8 9.3 6.0 43205.2 169443.9 5.5 392.8 838.6 5.0 5.0 43594.2 168161.4
6m-07 16.4 16.4 43200.2 14.8 715.8 18.0 14.8 43926.8 15.7 14.7 43206.8 167936.2 14.8 454.4 846.5 14.1 14.0 43671.9 170503.3
6m-08 9.2 7.8 43200.2 6.7 703.6 7.6 6.7 43907.2 7.2 7.2 43205.3 167703.6 6.9 552.2 1008.0 8.5 6.9 43714.8 169092.0
6m-09 4.8 3.7 43200.2 3.8 646.5 5.8 3.8 43836.1 3.7 3.7 43205.2 167972.9 3.8 525.3 909.9 4.3 3.8 43721.5 170618.2
6m-10 9.9 9.0 43200.2 9.0 1015.4 9.0 9.0 44143.2 10.0 9.2 43200.4 167217.1 9.0 722.2 1700.6 9.0 9.0 43905.5 171770.6
6m-11 6.9 6.1 43200.3 6.0 753.9 7.5 6.0 43927.7 6.0 6.0 43206.3 167171.3 6.0 696.3 1258.8 6.5 6.0 43914.7 170335.5
Strategic planning of an underground mine with variable cut‑off…
6m-12 8.2 8.0 43200.3 7.5 651.3 7.5 7.5 43766.2 10.1 8.6 43206.0 168879.6 7.4 579.8 1063.8 8.6 7.4 43770.6 171338.7
6m-13 18.6 17.0 43200.3 15.7 1594.3 17.2 15.7 44564.2 16.7 16.1 43207.9 168957.3 15.7 981.5 1963.9 15.7 15.7 44164.1 168789.7
6m-14 12.0 10.5 43200.2 8.8 803.7 8.8 8.8 44005.6 10.9 10.0 43206.4 165966.0 8.9 536.3 1032.7 10.4 8.9 43729.1 167389.7
6m-15 19.4 19.1 43200.2 16.7 1210.2 16.7 16.7 44524.9 19.6 18.1 43200.4 164884.2 16.8 581.1 1032.7 16.8 16.8 43785.2 165378.7
6m-16 16.5 16.5 43200.2 14.5 1508.9 14.8 14.5 44651.8 15.6 15.4 43200.3 162353.0 14.5 1614.1 3234.2 15.9 14.5 44624.0 167765.0
6m-17 16.7 16.7 43200.3 15.0 2242.3 15.4 15.0 45616.2 16.1 15.5 43208.4 165981.7 15.0 1382.8 3427.1 18.0 15.0 44535.2 168700.6
6m-18 15.3 15.3 43200.4 14.8 1001.2 14.9 14.8 44178.8 15.7 15.6 43200.5 165525.7 14.8 675.7 1365.1 16.3 14.8 43852.7 168172.4
6m-19 19.4 14.6 43200.3 12.9 1394.6 17.7 12.9 44586.2 17.3 13.4 43200.4 165416.1 12.7 1209.9 2732.7 12.7 12.7 44435.4 165504.2
6m-20 14.5 14.3 43200.2 17.3 2635.6 17.4 17.3 45906.3 15.6 14.4 43200.4 160889.3 18.3 2015.9 5027.9 13.5 13.5 45086.9 170094.7
6m-21 18.5 18.5 43200.4 15.6 1467.5 16.3 15.6 44768.5 19.8 18.7 43200.5 158846.6 15.6 890.1 1805.1 17.5 15.6 44088.4 161983.1
6m-22 17.8 17.5 43200.3 14.1 1164.4 14.1 14.1 44244.0 17.4 15.8 43200.5 163896.1 14.1 972.4 1922.1 15.1 14.1 44204.4 164518.4
6m-23 18.2 18.2 43200.4 13.3 1656.8 13.5 13.3 44883.2 17.7 16.4 43200.3 163490.7 13.3 1246.9 2900.9 14.5 13.3 44414.1 170183.6
13
Table 15 (continued)
Inst One thread Four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
6m-24 17.8 17.8 43200.4 19.7 3151.1 20.2 19.7 46642.1 21.7 21.1 43200.4 161372.6 19.9 1877.9 4259.9 20.9 19.9 44647.6 168415.4
6m-25 18.3 18.3 43200.4 19.7 4279.5 20.2 19.7 47680.3 20.9 20.5 43200.5 160555.0 20.1 2483.5 6496.8 20.8 20.1 45641.4 172135.3
Avg 12.8 11.7 43200.3 10.8 1251.8 12.2 10.8 44467.6 12.3 11.4 43203.5 165962.7 10.8 881.8 1909.0 11.3 10.6 44058.3 168549.8
R. Martinelli et al.
Table 16 Results for 7-mine instances
Inst One thread Four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
7m-01 2.0 2.0 43200.2 2.0 238.8 3.2 2.0 43443.8 2.0 2.0 43206.3 170549.9 2.0 202.2 347.2 2.3 2.0 43405.4 169691.1
7m-02 11.0 6.5 43200.3 6.2 434.0 12.4 6.2 43628.8 9.9 6.5 43204.2 169274.8 6.2 297.3 550.5 6.0 6.0 43499.7 170300.9
7m-03 5.2 3.2 43200.2 3.0 302.5 4.7 3.0 43539.9 3.0 3.0 43200.3 168852.8 3.0 268.5 407.8 4.5 3.0 43472.4 170149.2
7m-04 11.4 10.0 43200.2 9.9 950.5 11.4 9.9 44127.2 10.8 10.8 43205.5 168482.4 9.9 650.4 1358.5 10.1 9.9 43863.0 171481.5
7m-05 25.7 25.4 43200.2 24.3 975.5 24.3 24.3 44206.3 27.4 25.2 43203.6 169468.2 24.3 664.7 1319.7 25.9 24.3 43841.1 168230.9
7m-06 11.9 8.5 43200.2 7.6 512.6 10.3 7.6 43728.3 8.7 7.6 43206.3 166581.5 7.6 417.1 746.7 7.6 7.6 43615.1 166267.7
7m-07 5.9 4.3 43200.3 3.7 455.7 5.9 3.7 43658.6 4.3 3.8 43206.9 168344.5 3.7 431.2 675.1 3.7 3.7 43603.2 168414.2
7m-08 10.0 9.8 43200.3 9.1 714.8 9.5 9.1 43872.2 9.2 8.7 43208.5 166634.9 9.4 841.9 1372.4 9.3 9.3 43710.9 168744.0
7m-09 8.5 8.0 43200.4 8.3 643.6 8.2 8.2 43843.6 8.8 8.6 43206.3 166567.6 8.2 588.3 1032.8 8.7 8.2 43798.9 170333.6
7m-10 20.9 20.4 43200.3 18.5 1105.5 18.4 18.4 44469.3 20.4 19.3 43207.1 163668.9 18.5 694.1 1393.0 21.0 18.4 43841.1 169259.0
7m-11 19.2 17.3 43200.3 15.4 622.5 15.4 15.4 43844.0 19.2 15.5 43200.4 160969.3 15.4 544.4 925.0 16.8 15.4 43695.2 166105.1
Strategic planning of an underground mine with variable cut‑off…
7m-12 11.1 9.0 43200.3 8.9 1050.3 10.8 8.9 44239.6 14.6 11.1 43206.3 167023.9 8.9 823.2 1624.0 8.9 8.9 44006.1 165837.4
7m-13 6.5 6.5 43200.3 6.3 729.8 6.4 6.3 43963.2 7.5 7.5 43207.2 168130.2 6.3 637.1 1110.6 6.4 6.3 43805.3 167254.2
7m-14 10.5 10.5 43200.3 12.1 1053.1 14.0 12.0 44314.7 10.2 10.0 43206.9 163906.6 12.0 838.9 1454.6 10.5 8.6 44040.7 169808.6
7m-15 17.7 17.7 43200.4 13.5 909.6 13.7 13.5 44146.8 15.7 15.0 43200.4 164417.4 13.5 853.1 1592.0 14.3 13.5 44006.0 167172.3
7m-16 13.9 13.4 43200.3 11.9 991.8 11.9 11.9 44222.1 13.5 12.2 43200.4 166353.9 11.9 941.4 2060.4 12.3 11.9 44127.7 169854.7
7m-17 14.3 12.7 43200.3 10.4 1262.3 12.2 10.4 44535.5 13.5 11.6 43203.7 163816.5 10.4 1402.0 3082.8 10.4 10.4 44337.2 163231.0
7m-18 13.6 10.6 43200.3 9.6 1901.1 11.1 9.6 45224.0 11.0 9.6 43200.5 167052.2 9.6 1096.0 2284.3 9.6 9.6 44218.7 165170.9
7m-19 21.2 20.2 43200.3 17.9 835.4 18.0 17.7 44000.2 21.8 20.3 43206.8 161846.8 17.9 639.1 1249.4 17.7 17.7 43811.7 167609.2
7m-20 16.6 16.3 43200.3 13.3 2311.9 13.3 13.3 45600.0 17.0 15.8 43200.5 161436.6 14.2 1205.7 2600.0 14.7 14.1 44342.5 159735.8
7m-21 17.4 15.4 43200.4 13.9 1333.9 14.0 13.9 44546.4 16.2 14.7 43200.5 166387.6 13.9 1440.4 2515.4 13.9 13.9 44589.2 169994.5
7m-22 20.8 20.7 43200.4 21.9 2350.7 21.9 21.9 45618.2 22.9 22.4 43200.5 160808.7 21.5 2078.9 4617.6 21.8 21.5 45260.5 166986.9
7m-23 20.5 20.3 43200.4 19.5 1593.7 19.5 19.5 44953.9 20.5 18.9 43200.4 157681.3 19.5 1413.3 2615.4 20.6 19.5 44605.4 163027.1
13
Table 16 (continued)
Inst One thread Four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
7m-24 16.8 16.7 43200.3 15.9 4614.8 16.2 15.9 47568.6 10.6 10.5 43200.5 163669.2 15.7 3432.7 8135.9 15.7 15.7 46679.7 172512.4
7m-25 16.0 14.6 43200.3 12.6 4215.5 14.3 12.6 47158.9 13.7 11.7 43200.5 160741.0 12.4 3670.3 7593.4 12.4 12.4 47104.3 164746.7
Avg 13.9 12.8 43200.3 11.8 1284.4 12.8 11.8 44498.2 13.3 12.1 43203.6 165306.7 11.8 1042.9 2106.6 12.2 11.7 44211.2 167676.8
R. Martinelli et al.
Table 17 Results for 8-mine instances
Inst one thread four threads
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
8m-01 7.7 5.1 43200.2 4.8 378.3 8.5 4.8 43599.4 4.8 4.8 43209.7 168757.9 4.8 316.1 595.1 5.1 4.8 43527.1 169393.7
8m-02 11.7 10.8 43200.3 11.3 534.6 11.9 11.3 43766.0 12.9 12.9 43200.3 167273.0 11.3 541.1 934.6 11.3 11.3 43748.6 167794.1
8m-03 41.5 28.8 43200.3 24.9 456.8 36.4 24.9 43654.1 32.1 24.9 43204.2 169006.0 24.9 350.5 722.6 24.9 24.9 43571.4 169762.6
8m-04 17.4 12.4 43200.3 12.3 461.6 14.7 12.2 43625.1 12.6 12.6 43206.3 167203.9 12.3 370.6 696.5 12.9 12.3 43587.7 166629.4
8m-05 28.2 28.2 43200.3 27.5 1162.3 28.2 27.4 44332.5 29.1 27.8 43205.8 167694.1 27.5 689.5 1633.9 29.3 27.5 43910.3 168936.0
8m-06 16.1 13.8 43200.2 13.5 1384.7 15.5 13.5 44651.1 19.4 17.7 43205.0 164241.9 12.9 1188.9 2461.3 12.9 12.9 44431.3 157786.2
8m-07 13.0 10.8 43200.2 10.8 569.5 12.7 10.8 43771.8 12.0 12.0 43206.9 168070.0 10.8 594.5 1052.7 11.2 10.8 43791.4 170182.7
8m-08 23.4 18.1 43200.3 17.9 694.2 22.5 17.8 43929.2 24.4 18.8 43206.4 168057.2 17.9 580.8 1035.4 17.8 17.8 43802.9 166050.1
8m-09 24.3 20.8 43200.4 19.7 885.9 22.3 19.7 44207.0 22.5 19.7 43206.3 166585.7 19.7 725.3 1264.0 19.7 19.7 43934.7 168722.2
8m-10 15.8 15.5 43200.3 14.2 781.7 14.2 14.2 43984.4 14.9 14.8 43205.9 168498.1 14.2 742.7 1476.5 14.8 14.2 43944.7 171444.2
8m-11 15.0 13.4 43200.3 11.1 1310.2 11.7 11.1 44464.6 14.1 13.0 43207.6 167486.8 11.1 1225.3 2184.6 11.0 11.0 44280.5 169905.1
Strategic planning of an underground mine with variable cut‑off…
8m-12 25.8 21.5 43200.4 19.8 1361.7 24.1 19.8 44455.3 25.1 20.4 43209.4 164641.3 19.8 1200.1 1982.5 19.8 19.8 44287.3 165313.5
8m-13 15.8 14.8 43200.3 20.0 2287.1 17.9 17.4 45487.8 15.8 13.9 43200.6 164621.0 19.9 1878.4 2768.7 19.9 19.9 45129.6 169122.1
8m-14 17.7 17.7 43200.3 15.7 1417.2 15.9 15.7 44517.3 22.6 20.3 43200.6 161654.4 15.6 1031.6 1953.7 16.4 15.6 44256.2 168534.5
8m-15 17.9 17.6 43200.4 17.2 1437.3 17.2 17.2 44592.8 20.2 18.9 43205.5 164143.8 17.2 897.1 1793.4 17.2 17.2 44102.4 164923.4
8m-16 16.6 15.8 43200.4 13.3 1313.1 13.3 13.3 44476.6 15.4 14.7 43208.7 164139.7 13.4 1089.7 2068.2 13.7 13.4 44328.0 166241.5
8m-17 18.0 18.0 43200.4 21.4 2337.5 21.6 21.4 45537.4 17.3 16.6 43208.7 165619.4 21.3 2442.2 4742.4 18.4 17.9 45485.8 169314.9
8m-18 25.1 24.8 43200.3 16.6 3183.8 17.3 16.6 46309.7 21.7 20.4 43200.4 162490.4 16.6 2361.8 3653.5 16.6 16.6 45524.1 167199.2
8m-19 25.0 25.0 43200.3 22.4 2247.6 22.5 22.4 45254.0 24.9 23.3 43200.5 162814.4 22.0 1731.4 3413.3 22.9 22.0 44982.5 163189.5
8m-20 18.4 15.3 43200.3 10.9 1032.7 10.9 10.9 44287.5 14.9 13.5 43200.5 164218.7 10.9 1165.4 1896.4 13.2 10.9 44392.6 167660.9
8m-21 26.3 26.3 43200.2 17.0 2832.2 17.0 17.0 46055.5 20.7 20.6 43200.5 160811.8 16.7 3090.8 4869.9 17.5 16.7 46336.6 170033.4
8m-22 21.3 21.2 43200.4 27.2 6156.0 27.2 27.2 49616.7 22.5 22.3 43200.6 162768.0 28.1 5019.7 9830.8 22.8 22.8 48319.0 173643.8
8m-23 31.6 31.6 43200.4 16.9 3044.0 16.9 16.9 46428.7 25.2 24.2 43200.7 153010.1 17.9 2003.9 3965.0 19.0 17.9 45443.6 161971.4
13
Table 17 (continued)
Inst one thread four threads
13
Gown Gbest Twall Gbest Twall Gown Gbest Twall Gown Gbest Twall Tcpu Gbest Twall Tcpu Gown Gbest Twall Tcpu
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
8m-24 26.2 25.6 43200.3 18.6 3097.2 18.6 18.6 46404.8 23.8 22.8 43200.6 147755.5 17.7 1988.7 3785.2 18.3 17.7 45304.2 162652.2
8m-25 14.9 14.9 43200.6 10.7 2371.5 11.1 10.7 45588.2 13.4 13.1 43200.6 159073.9 10.8 2092.3 4244.6 11.7 10.8 45349.9 169542.2
Avg 20.6 18.7 43200.3 16.6 1709.5 18.0 16.5 44919.9 19.3 17.8 43204.1 164025.5 16.6 1412.7 2601.0 16.7 16.3 44630.9 167438.0
R. Martinelli et al.
Strategic planning of an underground mine with variable cut‑off…
References
Bley C, Boland N, Fricke C, Froyland G (2010) A strengthened formulation and cutting planes for the
open pit mine production scheduling problem. Comput Oper Res 37:1641–1647
Caccetta L, Hill SP (2003) An application of branch and cut to open pit mine scheduling. J Global Optim
27(2–3):349–365
Dagdelen K, Kuchta M, Topal EM, Moreno E, Enrique R (2002) Linear integer programming model applied
to scheduling of iron ore production at the kiruna mine. Trans Soc Min Metall Explor 312:194–198
Epstein R, Goic M, Weintraub A, Catalan J, Santibanez P, Urrutia R, Cancino R, Gaete S, Aguayo A,
Caro F (2012) Optimizing long-term production plans in underground and open-pit copper mines.
Oper Res 60(1):4–17
Gintner V, Kliewer N, Suhl L (2005) Solving large multiple-depot multiple-vehicle-type bus scheduling
problems in practice. OR Spectr 27(4):507–523
Hansen P, Mladenović N (2001) Variable neighborhood search: principles and applications. Eur J Oper
Res 130(3):449–467
Horsley T (2005) Differential cut-off grades. In: 9th AusIMM Underground Operators’ Conference, pp
103–109
Kuchta M, Newman A, Topal E (2004) Implementing a production schedule at lkab’s kiruna mine. Inter-
faces 34(2):124–134
L’Heureux G, Gamache M, Soumis F (2013) Mixed integer programming model for short term planning
in open-pit mines. Min Technol Trans Inst Min Metall A 122(2):101–109
Little J, Knights P, Topal E (2013) Integrated optimization of underground mine design and scheduling. J
South Afr Inst Min Metall 113(10):775–785
Martinez MA, Newman AM (2011) A solution approach for optimizing long and short-term production
scheduling at lkab’s kiruna mine. Eur J Oper 211:184–197
Nehring M, Topal E (2007) Production schedule optimisation in underground hard rock mining using mixed
integer programming. In: Australasian Institute of Mining and Metallurgy Publication Series, pp 169–175
Nehring M, Topal E, Kizil M, Knights P (2012) Integrated short-and medium-term underground mine
production scheduling. J South Afr Inst Min Metall 112(5):365–378
Newman AM, Rubio E, Caro R, Weintraub A, Eurek K (2010) A review of operations research in mine
planning. Interfaces 40(3):222–245
Pochet Y, Wolsey LA (2006) Production planning by mixed integer programming. Springer, New York
Pritsker A, Watters L, Wolfe P (1969) Multi-project scheduling with limited resources: zero-one pro-
gramming approach. Manag Sci 16:93–108
Rahal D, Smith M, Van Hout G, Von Johannides A (2003) The use of mixed integer linear programming
for long-term scheduling in block caving mines. In: Proceedings of the 31st International APCOM
Symposium, Cape Town, South Africa, pp 123–131
Riff MC, Otto E, Bonnaire X (2009) A new strategy based on grasp to solve a macro mine planning. In:
Rauch J, Ras Z, Berka P, Elomaa T (eds) Foundations of Intelligent Systems, vol 5722. Lecture
Notes in Computer Science. Springer, Berlin, pp 483–492
Sarin S, West-Hansen J (2005) The long-term mine production scheduling problem. IIE Trans
37(2):109–121
Smith M, O’Rourke A (2005) The connection between production schedule and cut-off optimization in
underground mines. In: Proceedings of the 32nd International APCOM Symposium, Tucson, Ari-
zona, 2005. A.A.Balkema, Leiden, pp 643–654
Smith M, Sheppard I, Karunatillake G (2003) Using mip for strategic life-of-mine planning of the lead/
zinc stream at mount ISA mines. In: Proceedings of the 31st International APCOM Symposium,
Cape Town, South Africa, pp 465–474
Terblanche S, Bley A (2015) An improved formulation of the underground mine scheduling optimisation
problem when considering selective mining. ORiON 31(1):1–16
Wang Q, Gu X, Chu D (2008) A dynamic optimization method for determining cutoff grades in under-
ground mines. Gospodarka Surowcami Mineralnymi 24(4/2):133–142
Wolsey LA (1998) Integer programming. Wiley, New York
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
13