IMPLEMENTATION OF LINEAR PROGRAMMING MODEL

FOR OPTIMUM OPEN PIT PRODUCTION SCHEDULING PROBLEM

By Edgar Urbaez and Kadri Dagdelen
Colorado School of Mines
Golden, Colorado 80401
ABSTRACT
Production scheduling is one of the most important factors affecting mine
planning. A program - Mine Scheduler - that aids mining engineers with the
setup of an optimum open pit production scheduling algorithm has been
developed. A mathematical model for open pit scheduling that optimizes the Net
Present Value of the cash flows was formulated. A simulated data is formulated
as a Mixed Integer Linear Programming (MILP) problem. Wha.t's Best by Lindo
Systems is used as the MILP solver in Microsoft Excel. The implementation of
the algorithm focuses on problem setup, allowing users to easily evaluate,
simplify, and develop the model that best fits their production scheduling.
INTRODUCTION
The objective of production scheduling is to try to best answer the
questions of whether a portion of a deposit should be mined or not, and if it is to
be mined when it should be mined and how it should be processed. Many
approaches have been taken in this area. However, due to the complexity of the
problem, in which sequencing of the pushbacks is one of the critical issues, it has
been difficult to come up with an algorithm that can truly produce an optimum
production schedule. Also, as complex as the algorithm is the manipulation of
the data. Mines nowadays are getting larger. The number of metallurgical
processes involved in a single operation has been increasing lately. The
optimum open pit production scheduling algorithm needs predefined pushbacks
before it can be used. As a consequence, in order to obtain the best results
possible from its implementation, the right open pit pushback design algorithm
has to be used. Due to the formulation of the algorithm, optimum cutoff grade
per period of time is achieved and different costs and prices can be included for
the life of a project.
PREVIUOS WORK
Different methods to solve the production scheduling problem have been
studied. Asarco's needs for a production scheduler (Williams, 1974) resulted in
the contracting of Systems Control, Inc. (SCI) to develop a computer program
that would solve the production scheduling problem one period at a time. Then,
through dynamic programming, the periods were combined to create the final
production schedule. Later, a linear programming formulation by Gershon
(Gershon, 1982) introduced the concept of long/short term interfaces. These
interfaces consisted of coming up with a short term schedule while keeping an
optimal long term plan. A new algorithm, based on the lagragian concepts of
mathematics, was then developed by Dagdelen and Johnson (Dagdelen and
Johnson, 1986) as an attempt to create a production schedule that was optimum.
The concepts of sequencing and multiple periods scheduling were successfully
introduced in this approach.
Since the mathematical programming approaches developed to address
the scheduling problem were difficult to implement and understand, Gershon's
heuristic approach seemed to yield a nearly optimal solution within a reasonable
period of time by utilizing the concept of a ranked positional weight to determine
whether a block should be mined or not (Gershon, 1987). This approach lacked
rigorous mathematical proof of the optimality of the solutions. A dynamic
programming algorithm was then formulated by Seymour (Seymour, 1994) that
finds the mining and cutoff grade sequence that maximizes the Net Present
Value. Since the algorithm tried to find a solution by exhaustive searching
techniques, it was prone to combinatorial explosion. Given that price of metal is
not constant, Wang and Sevim (Wang and Sevim, 1995) utilized Gershon's
downward cone concept to derive a method that is not a function of price but of
maximum-metal content as an alternative to parametrization.
In more recent works, techniques such as Gershon's heuristic, parametric
analysis, Wang and Sevim's heuristic, MILP, and exhaustive search via dynamic
programming, among others, have been developed. The main problem facing
each of these techniques is to come up with an ultimate pit limit and the
production schedule required to achieve it in one formulation; not one
independent of the other. Also, the time it takes to solve the problem is critical
for a given technique to be useful and effective for the mining engineers doing
the mine planning.
PROBLEM FORMULATION
Definition of the Value Coefficient
The Value Coefficient determines the economic value of a material type in
a given increment and period of time. It is used by the algorithm to find the most
profitable solution based on Net Present Value. It is expressed in $/ton and
calculated by using the following formula:
(
1)
k 0 k.o * k,o * k,o k.o *
Ci,j,l,m ($/ton) = Hgi,j,/ rj,j'/ (P- S)] = l7l
j
,j,l,m - Pi.j.I,m} (l+d)m
Where. cttl,m is the discounted value of material coming from mine k.
sequence '. material type o. increment i. to be processed by j in
time period m
As such the superscripts k,o and subscripts ;,j,' used in the variables refer
to material type 0 in mine k from sequence '. increment ;. to be processed by j in
a given time period m. The other variables are:
k.o
gj,j,1
rk,o
i,j,1
p
S
k.o .
l7li ,j,I,m
k,o
Pi,j'/.m
d
Grade (units/ton) associated with material described
by the superscripts k. 0 and subscripts i. j, ,
Recovery (%) for the process j
Commodity Price ($/units)
Marketing cost ($/units)
Mining cost ($Iton)
Processing cost ($Iton)
Discount rate or accepted rate of return that could be
realized on similar alternative investments of
equivalent risk (Stermole and Stermole . 1993)
Mathematical formulation of scheduling problem
The scheduling problem is formulated as a mixed integer programming
problem by Dagdelen (Dagdelen. 1996) and is presented as follows:
The objective function is to maximize the Net Present Value by varying the
flow of tonnages from source to destination over multiple time periods. The
mathematical formulation attempts to solve the different source-destination
combinations found in real life problems as represented in Figure 1. The
mathematical formulation is:
K I J L M N 0
Max Z - )
- .l-.i J..i "- "- "- !,J,I,m !,j.I,m !,I,m,n !,I,m,n I,J,m,n !,j,m,n
=1 1=1 j=1 =1 m=1 n=1 0=1
Let define the individual summations as:
K,l,J ,L,M ,N ,0
Max Z = + )
. . I,j,/,m 1t},I,m ',[,m,n ,,/,m,n I,},m,n l.j,m,n
k,I,j, ,nt,n,0=1
Where,
... ,,'

Nil <ill
Nil S (12)
'::::-""""'"'''":c .t5fJf> ...... ..a. (13)
Figure 1 Source-Destination diagram
K is the total number of mines
--.t5fJf> Hoof> 1_
:l iU'4> us>
I is the total number of increments in a sequence unit
J is the total number of processing techniques
L is the total number of sequences
M is the total number of time periods
N is the total number of increments in stockpiles
o is the total number of material types
is the discounted value coefficient per ton for material 0 in
mine k from sequence /, increment i, to be processed by j in
time period m
z;'tm,n is the discounted value coefficient per ton for material 0 in
mine k from sequence /, increment i, sent to stockpile n in
time period m
A;:j,m,n is the discounted value coefficient per ton for material 0 in
stockpile n in increment ito be processed by technique j
tUI,TnIII is the tonnage coming from mine k, material 0 in increment i,
from sequence / to be processed by technique j in time
period m
is the tonnage coming from mine k, material 0 in increment i,
l,i,m,n
from sequence / to be sent to stockpile n in time period m
itt},m,n is the tonnage coming from stockpile n, material 0 in
increment i, to be processed by technique j in time period m
Let simplify the summations from 1 to K,I,J,L,M,N,O as:
K .f.J .f-.M .N.O
=
Therefore, the objective function can be written as:
MaxZ= (C
k
,? +Zk.o st
k

o
+Ao it? )
. ',j.I,m l,j.I,m 1.I,m.n 1.I.m.n I.j.m.n I.j,m,n
k ... j m,n,o
Tonnage from mine k and material type 0 in increment i from a sequence I
can go to any destination j in any given time period as shown in Figure 2. This
tonnage is represented by the variable tUI,m' The objective function is subject to
the following constraints:

1I1l' a>
1111 t '"
IIU I
IItt,(Q
Figure 2 Variable Definition
1. Deposit reserves: ensures that material mined is what is available in the
geological reserves. Otherwise, the algorithm could allocate a tonnage of
material to a given value coefficient that exceeds what physically can be
found in the actual deposit, so that it can increase the resulting Net
Present Value. is the tonnage available for material 0 in mine k from
sequence " increment i. For k = 1, K; 0 = 1, 0; i = 1, I; I = 1, L; n = 1, N.
(t
k
.
o
+ stk,O ) $; Tk,o
i,j,l,m i,l,m,n i,l
j,m
2. Source mine production capacity: sets the mining rate. In other words, it
defines the minimum and maximum tonnage that can be mined from mine
k in time period m. For k = 1, K; m = 1, M.
k ( ko ko) M k
M mmm fi.J,/.m + sti.i.m.n maxm
'.J .. n.o
3. Destination processing capacity: sets the processing rate per destination.
It defines the minimum and maximum tonnage that can be sent to
destination j in time period m. For j = 1, J; m = 1, M.
C min j.m l (tU,.m + C max j.m
k ,fJ.:..o
4. Commodity unit production by destination and time period: sets the
commodity production target in measured units, such as ounces and
pounds among others. It defines the minimum and maximum commodity
production units that can be sent to process j in time period m. G is
recovered grade by k,o,i"j,l. SG is assumed recovered grade for each
stockpile. For j = 1, J; m = 1, M.
P min i.m l [( tU, * CU,) + * sctj.n)] P max j.m
k .fJ.:..o
5. Destination average percentage minimum (two sets): sets the lower head
grade limit per destination. It defines, in percentage, the minimum head
grade requirement to be processed by j in time period m. For j = 1, J; m =
1, M.

[
tk.o * (G
k

o
_ G . 0) . 0 * (SG
k
.
o
_ G . 0 )] ..." 0
i.j.l,m i,1 lUln j,m + lti,j,m,n n lUln j,m c:.
k., .. n,o
6. Destination average percentage maximum (two sets): sets the upper head
grade limit per destination. It defines, in percentage, the maximum head
grade requirement to be processed by j in time period m. For j = 1, J; m =
1, M.
1 [tUI,m * - Gmaxj,m) + * - Gmaxj,m)] 0
k /;:';',0
7. Stockpile inventory: determines the availability of material in the stockpile.
It defines the tonnage available that can be sent to any destination from
the stockpile by keeping track of the current stockpiled material and the
new stockpiled material in a given time period. It:::.n is the stockpile
inventory consisting of material 0, from mine k, increment i, in time period
m, in stockpile n. For k = 1, K; 0 = 1,0; i = 1, I; m = 1, M; n = 1, N.
/k.o = / k.o + l s t ~ o _ itO )
'.m.n ,. m -I. n ~ ,.I.m.n I.j.m.n
j.
8. Integer check: ensures that only one integer has a value of one. If an
integer has a value of one, a sequence can start mining. If a sequence
can start mining, any previous sequence can be partially mined or
completely mined out. Y/
m
is one if cumulative tons from mine k, sequence
I are depleted in period m or before. Y/
m
is zero otherwise. For k = 1, K; I =
1, L.
9. Integer conditioning: activates the mining of a sequence in a given period
of time. T/ is the total tonnage in sequence Ifrom mine k. For k = 1, K; 1=
1, L-1; mm = 1, M.
M M
E E tUI.mm - [( E Y/mm ) * T/] ~ 0
mm-l l.j,O mm=I
10. Sequence enforcing: ensures that a sequence has to be completely mined
out before mining material from a third sequence. This constraint allows
for partial mining, which means that two sequences can be mined at the
same time, but a third sequence can not start mining until the first
sequence is all mined out. For k = 1, K; I = 1, L; m = 1, M.
~ k.o (Y k * T k ) " 0
/.J ti.j,/+I.m - I.m 1+1::::'
'.j.O
SIMULATED MODEL
Setting up the scheduling problem
The simulated data is designed to convey the output information in a way
that is easy to access by the user. The assumptions are realistic and the data
represents a real life gold deposit. The simulated model consists of 1 source, 3
processes (1 mill, 1 heap pile, 1 dump site), 2 material types, 3 pushbacks, 5
material increments, 3 time periods and 1 stockpile resulting in 370 numeric
variables as shown in Figure 3.
Cost parameters are setup in Microsoft Excel by Mine Scheduler as
shown in Figure 4. The light blue cells represent the input fields. As can be
seen, cost can incrementally increase as the pit gets deeper. Also, none of the
parameters has to be assumed constant throughout the mine life. Net Present
Value formula is included in this model. Discounting starts at the end of a year.
Costs for stockpiling and rehandling are also included.
ECONOMIC PARAMETERS
Discount Rate
Gold Price ($/oz)
Sales Cost ($/oz)
Stockpile Cost ($/ton)
Rehandle Cost ($/ton)
Process Cost :
Mill - Oxide ($/ton)
- Sulfide($/ton)
Leaching - Oxide ($/ton) ' ~ ~ ~ i I ~ & ~ ~
- Sulfide($/ton) 14.
Mining Cost :
PushBack 1 ($Iton)
Push Back 2 ($/ton)
PushBack 3 ($/ton)
Waste Cost
Figure 4 Cost parameters table
This model contains 2 different material types: oxide and refractory. Figure
5 shows the reserve table for the refractory and oxide material types. Both
material types can be defined in terms of grade increments as well as tonnage
and average grade within these increments. Refractory material can be further
defined in terms of sulfur/sulfide (SS) and carbon/carbonate (CC) content. The
average grade is determined by calculating the weighted average of the grade
increment.
Constraints are setup for the different destinations as shown in Figure 6.
Large numbers are inputted as maximum limits for Stockpile and Waste Dump to
simulate unlimited capacities. Stockpile constraints are removed from the last
year since the objective is to handle and process all the material available by the
end of the mine life. There is a minimum annual production of 20,000,000 tons
from the mine. Also, the mill has to have an annual supply of at least 7,000,000
tons with a minimum average grade of 0.04 gold ounces per ton.
PB
Figure 5 Resource table
Mine, Tons max.
Tons min.
Mill Tons max.
Leaching
Waste Dump
Stockpile
Tons min.
Grade max.
Grade min.
SS content
Tons max.
Tons min.
Grade max.
Grade min.
SS content max .....

SS content min
CC content
Tons min.
Tons max.
Tons min.
Figure 6 Constraint table
Recovery tables are created for each material type. Figure 7 shows the
recovery tables for both material types. Recovery values are inputted for each
process. These values do not need to be constant. They can vary by pushback
and also by grade increment within a pushback. This approach results in a more
powerful and flexible tool as more detailed information coming from a grade-
recovery curve can be inputted.
OXIDE
PushBack 3 ~ ~ ~ ~
Figure 7 Recovery table
OXIDE -1998
PushBack 1
PushBack 2 ~ ~ ~ ~ __ - - : ~ __ ~ _ ~ ~ _ . . , . . . . . . : ~
Figure 8 Year 1998 input solution table
Variables are created to represent the tonnage that is mined from a given
year, material type, pushback, and material increment. Figure 8 shows a solution
table for the year 1998. A similar table is created for each period of time and for
each material type. The resulting output has the same table format as the input
data, thus providing an easy to read and report production scheduling
spreadsheet.
Solving the scheduling problem
Once the problem has been formulated and setup, solving it is as easy as
going to the WB! Button in Microsoft Excel and choosing Solve. Information
regarding the number of variables, number of constraints and type of model are
displayed in a status window as the MILP solver is searching for a solution.
Figure 9 shows this window right before it writes the globally optimal solution for
the scheduling problem.
Figure 9 Solver status window
The speed at which the MILP solver finds a solution depends on the
computer under which it is run. The faster the clock speed, the faster the
information is processed. The more random access memory (RAM's) there is,
the faster it can swap information between the software and the hard drive. The
effect of more RAM's is greatly noticeable when dealing with integer variables.
Analyzing the solution
After What's Best solved the production scheduling formulation, the
solution is posted on the Microsoft Excel spreadsheet as shown in Figure 10.
The maximum Net Present Value found is $ 338,523,408. As can be seen for
1998, oxide material is mined and sent to all different destinations, including the
stockpile. The classification of the material is done based on the Value
Coefficient assigned to a tonnage belonging to a given grade increment,
pushback, material type, time period and mine. For this model, the mine life is 3
years, or 3 time periods, as years can be grouped in a single time period.
Figure 10 Output solution tables
Figure 10 also shows that no refractory material is mined in 1998, thus the
total tonnage mined from pushback 1 is 15 millions. In 1999, another 15 millions
tons are mined from pushback 1, but 446,058 tons of refractory material are sent
to waste. Since pushback 1 has not been depleted by 1999, mining of pushback
2 and pushback 3 is not possible. Furthermore, 3,578,120 tons of oxide material
are sent to the stockpile in 1999.
CONCLUSION
A mathematical model for open pit scheduling problem that optimizes the
Net Present Value of the cash flows has been formulated as a Mixed Integer
Linear Programming problem. This work focuses on the implementation of the
open pit scheduling problem formulation. A program, Mine Scheduler, is
developed to setup the studied formulation. Mine Scheduler provides a fast and
reliable way to setup open pit production scheduling problems. It eliminates the
risks of accidentally typing the wrong formula in a cell. Its ability to relatively
quickly setup an open pit production scheduling algorithm makes for a powerful
tool as more scenarios can be evaluated per period of time compared to
traditional methods. Also, its Microsoft Windows interface makes it extremely
easy to use requiring no special training. Just a quick tutorial, taking place in few
minutes, is enough to start using Mine Scheduler.
Some of the advantages of using Mine Scheduler are the dynamic
determination of cutoff grades per destination and per time period, the non
constant yearly price and cost input permitting forecasting, and its flexibility of
use of the different constraint sets. Another advantage is the manipulation of the
final results, which can be exported in any text format and read by a variety of
text editors. Results can also be expressed in 2-D and 3-D graphics by using the
existing tools in Microsoft Excel.
REFERENCES
Dagdelen, K., 1985, "Optimum Multi Period Open Pit Mine Production
Scheduling", Ph.D. Dissertation, Colorado School of Mines, Golden, Colorado
Dagdelen, K. and Johnson, T., 1986, "Optimum Open Pit Mine Production
Scheduling by Lagrangian Parametrization", 19
TH
APCOM SYMPOSIUM, pp.
127-142
Dagdelen, K., 1996, "Formulation of Open Pit Scheduling Problem Including
Sequencing as MILP", Internal Report, Mining Engineering Department, Colorado
School of Mines, Golden, Colorado
Davis, R. and Williams, C., 1972, "Optimization Procedures for Open Pit Mine
Scheduling", pp. C1-C18
Gershon, M., 1982, "A Linear Programming Approach to Mine Scheduling
Optimization", 17TH APCOM SYMPOSIUM, pp. 483-499
Gershon, M., 1987, "An Open-Pit Production Scheduler: Algorithm and
Implementation", Mining Engineering, pp. 793-796
Sevim, H. and Lei, D., 1995, "A Dynamic-programming-based Algorithm for
Optimal Production Planning in Open Pit Mines", Society for Mining, Metallurgy,
and Exploration, inc., transactions vol. 296, pp. 1851-1855
Seymour, F., 1994, "Finding the Mining Sequence and Cutoff Grade Schedule
that Maximizes Net Present Value", draft
Stermole, F. and Stermole, J., 1993, "Economic Evaluation and Investment
Decision Methods" Eight Edition, Investment Evaluations Corporation, Golden,
Colorado, p. 12
Wang, Q. and Sevim, H., 1968, "Open Pit Production Planning through Pit-
generation and Pit Sequencing", Society for Mining, Metallurgy, and Exploration,
inc., Transactions vol. 294, pp. 1968-1974
Wang, Q. and Sevim, H., 1995, "Alternative to Parametrization in Findings a
Series of Maximum-metal Pits for Production Planning", Mining Engineering, pp ..
178-182
Wells, H., 1978, "Optimization of Mining Engineering Design in Mineral
Valuation", Mining Engineering, pp. 1676-1684
Williams, C., 1974, "Computerized Year-by-Year Open-Pit Mine Scheduling",
AIME, transactions vol. 256, pp. 309-317