See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225689475

Stochastic optimization for strategic mine planning: A decade of developments

Article  in  Journal of Mining Science · March 2011

DOI: 10.1134/S1062739147020018

CITATIONS READS

55 1,103

1 author:

Roussos Dimitrakopoulos
McGill University
213 PUBLICATIONS   2,695 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Stochastic Simultaneous Optimization of Mining Complexes - Mineral Value Chains View project

All content following this page was uploaded by Roussos Dimitrakopoulos on 04 January 2014.

The user has requested enhancement of the downloaded file.

Journal of Mining Science, Vol. 47, No. 2, 2011

STRATEGIC MINE PLANNING UNDER UNCERTAINTY

STOCHASTIC OPTIMIZATION FOR STRATEGIC

MINE PLANNING: A DECADE OF DEVELOPMENTS

R. Dimitrakopoulos UDC 539.3

Conventional approaches to estimating reserves, optimizing mine planning, and production forecasting
result in single, and often biased, forecasts. This is largely due to the non-linear propagation of errors in
understanding orebodies throughout the chain of mining. A new mine planning paradigm is considered
herein, integrating two elements: stochastic simulation and stochastic optimization. These elements
provide an extended mathematical framework that allows modeling and direct integration of orebody
uncertainty to mine design, production planning, and valuation of mining projects and operations. This
stochastic framework increases the value of production schedules by 25 %. Case studies also show that
stochastic optimal pit limits (i) can be about 15 % larger in terms of total tonnage when compared to the
conventional optimal pit limits, while (ii) adding about 1 0% of net present value to that reported above for
stochastic production scheduling within the conventionally optimal pit limits. Results suggest a potential
new contribution to the sustainable utilization of natural resources.

Mine planning, stochastic optimization, geological uncertainty, simulated annealing, production scheduling

INTRODUCTION
Optimization is a key aspect of mine design and production scheduling for both open pit and
underground mines. It deals with the forecasting, maximization and management of cash flows from a
mining operation, and is the key to the financial aspects of mining ventures. A starting point for
optimization in the above context is the representation of a mineral deposit in three-dimensional space
through an orebody model and the mining blocks representing it; this is used to optimize designs and
production schedules (e.g. [1]). Geostatistical estimation methods have long been used to model the
spatial distribution of grades and other attributes of interest within the mining blocks representing a
deposit [2]. The main drawback of estimation techniques, be they geostatistical or not, is that they are
unable to reproduce the in-situ variability of the deposit grades, as inferred from the available data.
Ignoring such a consequential source of risk and uncertainty may lead to unrealistic production
expectations (e.g. [3]). Figure 1 shows an example of unrealistic expectations in a relatively small gold
deposit. In this example [3], the smoothing effect of estimation methods generates unrealistic
expectations of net present value (NPV) in the mine’s design, along with ore production performance,
pit limits and so on. The figure shows that if the conventionally constructed open pit design is tested
against equally probable simulated scenarios of the orebody, its performance will probably not meet
expectations. The conventionally expected NPV of the mine has a 2 to 4 percent chance to materialize,
while it is expected to be about 25 % less than forecasted. Note that in a different example, the opposite
could be the case.

COSMO - Stochastic Mine Planning Laboratory, Department of Mining and Materials Engineering, McGill
University, E-mail: roussos.dimitrakopoulos@mcgill.ca, Montreal, Canada.
©
1062-7391/11/4702-0138 2011 Pleiades Publishing, Ltd.

138
 Fig. 1. Optimization of mine design in an open pit gold mine, NPV versus “pit shells” and risk profile of
the conventionally optimal design

For several decades now, a traditional framework has been established and used when dealing with
the modeling of the spatial distribution of attributes of a mineral deposit, optimization for mine design
and planning, to support downstream studies, valuation, and decision-making. Starting from the
beginning of this decade, a different framework than the traditional one has been pursued and is
outlined in Fig. 2. Instead of a single orebody model as an input to optimization for mine design and a
correct assessment of individual key project indicators, a set of equally probable models of the deposit
can be used. These models are conditional to the same available data and their statistical characteristics,
are constrained to reproduce all available information, and represent equally probable models of the
actual and unknown spatial distribution of grades [4]. The availability of multiple equally probable
models of a deposit enables mine planners to assess the sensitivity of pit design and long-term
production scheduling to geological uncertainty (e.g. [5, 6]) and, more importantly, empower mine
planners to produce mine designs and production schedules with substantially higher NPV, larger pit
limits, and longer LOM assessments through stochastic optimization. Figure 3 shows an example from
a major gold mine presented in Godoy and Dimitrakopoulos [7], where a stochastic approach leads to a
marked improvement of 28 % in NPV over the life of the mine, compared to the standard best practices
employed at the mine; note that the pit limits used are the same in both cases and are conventionally
derived through commercial optimizers [1]. The same study also shows that the stochastic approach
leads to substantially lower potential deviation from production targets, that is, reduced risk. A key
contributor to substantial differences is that the stochastic or risk-integrating approach can distinguish
between the “upside potential” of the metal content, and thus the economic value of a mining block,
from its “downside risk”, and then treat them accordingly, as further discussed herein.
Figure 2 represents an extended mine planning framework that is stochastic (that is, integrates
uncertainty) and encompasses the spatial stochastic model of geostatistics with that of stochastic
optimization for mine design and production scheduling. Simply put, in a stochastic mathematical
programming model developed for mine optimization, the related coefficients are correlated random
variables that represent the economic value of each block being mined in a deposit, which are in turn
generated from considering different realizations of metal content. This second key element of the risk-
integrating approaches is stochastic simulation and is further discussed below. The further integration
of market uncertainties in terms of commodity prices and exchange rates is discussed elsewhere [9, 10].

139
 Fig. 2. Traditional (deterministic or single model) view and practice versus risk-integrating (or stochastic)
approach to mine modeling, from reserves to production planning and life-of-mine scheduling, and
assessment of key project indicators

The key idea in production scheduling that accounts for grade uncertainty is relatively simple. A
conventional optimizer (any one of them) is deterministic by construction and evaluates a cluster of
blocks, such as that in Fig. 4a, so as to decide when to stop mining, which blocks to extract when, and
so on, assuming that the economic values of the mining blocks considered (inputs to the optimizer) are
the actual/real values. A stochastic optimizer, also by construction, evaluates a cluster of blocks but as
in Fig. 4b, by simultaneously using all possible combinations of economic values of the mining blocks
in the cluster being considered. As a result, substantially more local information on joint local
uncertainty is utilized, leading to much more robust schedules that also can maximize the upside
potential of the deposit (e.g. higher NPV and metal production) and at the same time minimize
downsides (e.g. not meeting production targets and related losses).
To elaborate on the above, the following paragraphs examine the key element in the risk-
integrating framework shown in Fig. 2, that of stochastic optimization. The next sections start from the
generation of a major input to the optimization process, the simulated representations of the deposit
considered. Stochastic optimization is presented in two approaches, one based on the technique of
simulated annealing, and a second based on stochastic integer programming. Examples follow that
demonstrate the practical aspects of stochastic mine modeling, including the monetary benefits.

Fig. 3. The stochastic life-of-mine schedule in this large gold mine has a 28 % higher value than the best
conventional (deterministic) one. All schedules are feasible
140
Conventional optimization:

Maximize ( с1x11 + с2 x12 + ... …. )

Subject to … ,

where с n is the economic value of a block (a

constant); xnp is a binary variable; mine or not
block n in period p.
The economic value of a block сn is: Stochastic optimization:
Maximize
(METAL*RECOVERY*PRICE- ( с11x11 + с12 x12 + ... + с12 x11 + с22 x12 + ... + с1r x1p + ... + сnr xnp )
ORE*COSTProc) – Subject to … ,
ROCK*COSTMin
where r is a given simulation of the deposit

Fig. 4. Production scheduling optimization with conventional versus stochastic optimizers. a — single
representation of a cluster of mining blocks in a mineral deposit as considered for scheduling by a
conventional optimizer; b — a set of models of the same cluster of blocks with multiple possible values
considered simultaneously for scheduling by a stochastic optimizer

MODELING UNCERTAINTY IN METAL SUPPLY: HIGH-ORDER SEQUENTIAL SIMULATION WITH

SPATIAL CUMULANTS
The sequential simulation approach employed to simulate pertinent attributes of mineral deposits is
briefly outlined in this section, and is followed by a generalization of the approach using high-order
statistics. This approach is suitable for the stochastic simulation of complex non-Gaussian spatial
attributes of deposits with complex non-linear geological patterns, and is based on the notion of high-
order spatial cumulants [11]. The approach requires the use of geological analogues or training images
(TI) to infer the high-order relations not available in the existing data. Advantages of the high-order
sequential simulation approach outlined here are the (a) absence of pre-processing steps, such as data
normalization or training image filtering; (b) use of high-order relations in the available data that
dominate the simulation process (i.e. data-driven, not TI-driven); (c) generation of complex spatial
patterns that reproduce bi-modal data distributions, data variograms and high-order spatial cumulants.
Consider a stationary and ergodic random field Z ( xi ) or Z i xi ∈ xi ∈ Ω ⊆ R n ( n = 1, 2 or 3 ) for
i = 0, ..., N where N is the number of points in a discrete grid D N and a set of conditioning data
d n = {Z ( xα ), α = 1, ..., m} . In addition, we introduce the set Λ i such that Λ 0 = {d n }
Λ i = Λ i +1 ∪ {Z ( xi −1 )} Let f z ( z0 , z1, ..., z N ) be a probability distribution function associated with a
141
multivariate process Z = {Z 0 , Z1 , ..., Z N } at x = {x0 , x1, ..., xN } . The multivariate distribution f Z can de
decomposed, based on the Bayes relation, into the product of univariate conditional distributions
functions [12, 13]:
f Z ( z | Λ 0 ) = f Z ( z 0 , z1 , ..., z N | Λ 0 ) = f Z ( z0 | Λ 0 ) f Z ( z1 | Λ 1 ) ... f Z ( z N | Λ N ) ,
0 1 N
(1)

where f Z i ( zi | Λ i ) is the conditional distribution of Zi , given Λi . To generate a realization

z = ( z0 , ..., z N ) of Z, a value z0 is drawn for Z 0 based on f Z 0 ( z 0 | Λ 0 ) ; then, for i = 1, ..., N ) , zi is
drawn from the conditional probability function f Z i ( zi | Λ i ) . For additional details, see Journel [4],
and Dimitrakopoulos and Luo [14]. The estimation of f Z in Eq. (1) given Λ 0 needs the estimation of
( N + 1 ) local conditional probability density functions (cpdfs). The cpdfs can be estimated using the
well-known sequential conditional simulation algorithm which follows the sequence: (1) randomly
choose the spatial location of a node xi to be simulated; (2) estimate f Z i ( zi / Λ i ) ; (3) draw a value zi
from f Zi ( z i / Λ i ) , which becomes a conditioning data for all subsequent drawings; (4) return to step (1)
until all nodes have been visited using a random path.
For the reason of simplicity and without loss of generality, assume that x0 is the first node visited
and its neighbours are found within a certain neighbourhood. The cpdf f Z 0 given Λ 0 is defined by:
f Z ( z)
f Z ( z0 | Λ 0 ) = , (2)
0
f Z ( z0 )
0

where z 0 = ( z1 ,..., zn ) Z0 = ( Z1 ,..., Z n ) and f Z 0 is the marginal density. It is possible to derive an

estimation of f Z 0 without transforming the conditioning data as, for example, made by the Gaussian-
related or indicator-based algorithms, through the high-dimensional Legendre polynomials combined
with high-order spatial cumulants used to derive analytical expressions to the local cpdfs f Z i ( zi | Λ i )
The cpdf f Z 0 can be written as:
ω iN − 2 iN −1
1
f Z0 ( z 0 | Λ 0 ) = ∑ ... ∑ ∑ Li ,i ,..., i N −1 ,iN
Pi0 ( z 0 ) , (3)
∫D f Z (x)dz 0
0 1
i =0 i =0 i =0
0 N −1 N

where Li0 ,..., iN −1 ,iN = Li0 ,..., iN −1 ,iN Pi1 ( z1 ) ... PiN −1 ( z N −1 ) PiN ( z N ) ; D is the working domain; Pm (z ) is the m th
order normalized Legendre polynomials; the Legendre cumulants are Lm = g m (ci ) , i = 0, ..., m ,
m = 0, 1, 2... , where ci is the ith-order (spatial) cumulant of f and are calculated from available data
and training image (TI) or analogue. Dimitrakopoulos et al. [11] and Mustapha and Dimitrakopoulos
[8, 15] detail the inference and interpretation of spatial cumulants and their relation to high-order
moments.
The high-order sequential simulation (HOSIM) is a general approach to stochastic simulation, and
follows the main steps below:
1. Scan the TI and sample data and store the spatial cumulants calculated, required by the Legendre
series in Eq. (3).
2. Define a random path visiting once and only once all unsampled nodes to be simulated.
3. Define a template for each unsampled location x0 using its neighbours. The conditioning data
available are then searched. The high-order spatial cumulants are read from the global tree in Step 1, and
are used to calculate the coefficients of the Legendre series and build the local cpdf of Z 0 using Eq. (3).
142
 4. Draw a uniform random value in [0,1] to read from the conditional distribution a simulated value
Z ( x0 ) at x0 .
5. Add x0 to the set of sample hard data and the previously simulated values.
6. Repeat Steps 3 and 5 for the next points in the random path defined in Step 2.
7. Repeat Steps 2 to 6 to generate different realizations using different random paths.
An example from the use of the above algorithm is presented in Fig. 5, which depicts a complete
image, samples from that image used for high-order simulation with spatial cumulants, and two
simulated realizations. Figure 6 shows fourth-order cumulant map (slices) of data (left), two
realizations shown in Fig. 5, and demonstrates that the reproduction of high-order spatial statistics is
excellent. Further details are available in [14]. Note that the simulation method above reduces to the
usual second-order simulation methods commonly used. The deposit descriptions used in the stochastic
optimization that follows in the next sections are based on the common “sub-case” of second-order
Gaussian simulation algorithms.
STOCHASTIC OPTIMIZATION IN MINE DESIGN AND PRODUCTION SCHEDULING

Mine design and production scheduling for open pit mines is an intricate, complex and difficult
problem to address due to its large scale and uncertainty in the key parameters involved. The objective
of the related optimization process is to maximize the total net present value of the mine plan. One of
the most significant parameters affecting the optimization is the uncertainty in the mineralized
materials (resources) available in the ground, which constitutes an uncertain supply for mine
production scheduling. A set of simulated orebodies provides a quantified description of the uncertain
supply. Two stochastic optimization methods are summarized in this section. The first is based on
simulated annealing [7, 16, 17]; and the second on stochastic integer programming [18 - 21].

Fig. 5. Complete image, samples used and two realizations from HOSIM; one with all terms used in
HOSIM (Legendre series approximation) and one without the terms discussed in the text

Fig. 6. Fourth-order cumulant map (slices) of data (left), first realization and second realization shown in
Fig. 4 — the reproduction of high-order stats is excellent
143
 Fig. 7. Steps needed for the stochastic production scheduling with simulated annealing; S1... Sn are
realizations of the orebody grade through a sequential simulation algorithm; Seq1 ... Seqn are the mining
sequences for each of S1... Sn. Mining rates are input to the process

Production Scheduling with Simulated Annealing. Simulated annealing is a heuristic

optimization method that integrates the iterative improvement philosophy of the so-called Metropolis
algorithm with an adaptive “divide and conquer” strategy for problem solving [22]. When several mine
production schedules are under study, there is always a set of blocks that are assigned to the same
production period throughout all production schedules; these are referred to as the certain or 100 %
probability blocks. To handle the uncertainty in the blocks that do not have 100 % probability,
simulated annealing swaps these blocks between candidate production periods so as to minimize the
average deviation from the production targets for N mining periods, and for a series of S simulated
orebody models, that is:
N ⎛ S S ⎞
MinO = ∑ ⎜⎜ ∑ θ n*( s ) − θ n ( s ) + ∑ ωn* ( s ) − ωn ( s ) ⎟⎟ , (4)
n =1 ⎝ s =1 s =1 ⎠
where θ n* ( s ) and ωn* ( s ) are the ore and waste production targets, respectively, and θ n (s ) and ωn (s )
represent the actual ore and waste production of the perturbed mining sequence. Each swap of a block
is referred to as a perturbation. The probability of acceptance or rejection of a perturbation is given by:
⎧1, if Onew ≤ Oold ,
⎪
Pr ob{accept} = ⎨ Oold −Onew
⎪⎩e T , otherwise.
This implies all favorable perturbations ( Onew ≤ Oold ) are accepted with probability 1 and
unfavorable perturbations are accepted based on an exponential probability distribution, where T
represents the annealing temperature.
The steps of this approach are depicted in Fig. 7 and are: (a) define ore and waste mining rates; (b)
define a set of nested pits as per the Whittle implementation [1] of the Lerchs-Grossmann [23]
algorithm or any pit parameterization; (c) use a commercial scheduler to schedule a number of
simulated realizations of the orebody given (a) and (b); (d) employ simulated annealing as in Eq. (1)
using the results from (c) and a set of simulated orebodies; and (e) quantify the risk in the resulting
schedule and key project indicators using simulations of the related orebody.
STOCHASTIC INTEGER PROGRAMMING FOR MINE PRODUCTION SCHEDULING
Stochastic Integer Programming for Mine Production Scheduling. Stochastic integer
programming (SIP) provides a framework for optimizing mine production scheduling considering

144
uncertainty [24]. A specific SIP formulation is briefly shown here that generates the optimal production
schedule, using equally probable simulated orebody models as input, without averaging the related
grades. The optimal production schedule is then the schedule that can produce the maximum achievable
discounted total value from the project, given the available orebody uncertainty described through a set
of stochastically simulated orebody models. The proposed SIP model allows the management of
geological risk in terms of not meeting planned targets during actual operation. This is unlike the
traditional scheduling methods that use a single orebody model, where risk is randomly distributed
between production periods while there is no control over the magnitude of the risks on the schedule.
The general form of the objective function is expressed as”
p
⎡n m
tg ⎤
Max ∑ ⎢∑ E{(NPV) it }bit − ∑ ( cuto d suto + cmto d sm
to
+ cutg d sutg + cmtg d sm )⎥ , (5)
t =1 ⎣ i =1 s =1 ⎦
where p is the total production periods, n is number of blocks, and bit is the decision variable for when
to mine block i (if mined in period t, bit is 1 and otherwise bit is 0). The c variables are the unit costs of
deviation (represented by the d variables) from production targets for grades and ore tonnes. The
subscripts u and l correspond to the deviations and costs from excess production (upper bound) and
shortage in production (lower bound), respectively, while s is the simulated orebody model number,
and g and o are grade and ore production targets. Figure 8 graphically shows the second term in Eq. (5).
Note that the cost parameters in Eq. (5) are discounted by time using the geological risk discount
factor developed in Dimitrakopoulos and Ramazan [25]. The geological risk discount rate (GRD) allows
the management of risk to be distributed between periods. If a very high GRD is used, the lowest risk
areas in terms of meeting production targets will be mined earlier, and the most risky parts will be left for
later periods. If a very small GRD or a GRD of zero is used, the risk will be distributed at a more
balanced rate among production periods, depending on the distribution of uncertainty within the
mineralised deposit. The c variables in the objective function (Eq. (5) are used to define a risk profile for
the production, and the NPV produced is the optimum for the defined risk profile. It is considered that if
the expected deviations from the planned amount of ore tonnage having planned grade and quality in a
schedule are high in actual mining operations, it is unlikely to achieve the resultant NPV of the planned
schedule. Therefore, the SIP model contains the minimization of the deviations together with the NPV
maximization to generate practical and feasible schedules and achievable cash flows. For details please
see Ramazan and Dimitrakopoulos [19] and Dimitrakopoulos and Ramazan [24].

Fig. 8. Graphic representation of the way the second component of the objective function in Eq. (2)
minimizes the deviations from production targets while optimizing scheduling. This leads to schedules

145
 where the potential deviations from production targets are minimized, leading to schedules that seek to
mine first not only for high grade mining blocks, but also with high probability to be ore

Fig. 9. a — risk-based LOM production schedule (cumulative NPV risk profile); b — risk-based LOM
production schedule (ore risk profile); c — risk-based LOM production schedule (waste risk profile)

EXAMPLES AND VALUE OF THE STOCHASTIC FRAMEWORK

The example discussed herein shows long-range production scheduling with both the simulating
annealing approach in Section 3.1 and SIP model in Section 3.2. Section 3.3 focuses on the topic of
stochastically optimal pit limits. The application used is at a copper deposit comprising 14 480 mining
blocks. The scheduling considers an ore capacity of 7.5 Mt per year and a maximum mining capacity is
28 Mt. All results are compared to the industry’s “best practice”: a conventional schedule using a single
estimated orebody model and Whittle’s approach [1].
Simulated Annealing and Production Schedules. The results for simulated annealing and the
method in Eq. (1) are summarized in Fig. 9 to 11. The risk profiles for NPV, ore tonnages, and waste
production are respectively shown in Fig. 9. Figure 10 shows a comparison with the equivalent best
conventional practice, and reports a difference of 25 % in terms of higher NPV for the stochastic
approach.
SIP and Production Schedules. The application of the SIP model in Eq. (2), using pit limits
derived from the conventional optimization approach, forecasts an expected NPV at about $238 M.
When compared to the equivalent traditional approach and related forecast, the value of the stochastic
framework is $60 M, or a contribution of about 25 % additional NPV to the project. Note that unlike
146
simulated annealing, the scheduler decides the optimal waste removal strategy, which is the same as the
one used in the conventional optimization with which we compare.

Fig. 10. NPV of conventional and stochastic (risk-based) schedules and corresponding risk profiles

Figure 11 shows a cross-section of the two schedules from the copper deposit; one obtained using
the SIP model (bottom) and the other generated by a traditional method (top) using a single estimated
orebody model. Both schedules shown are the raw outputs and need to be smoothed to become
practical. It is important to note that (i) the results in the second case study are similar in a percentage
improvement when compared to other stochastic approaches such as simulated annealing, and (ii)
although the schedules compared in the studies herein are not smoothed out, other existing SIP
applications show that the effect of generating smooth and practical schedules has marginal impact on
the forecasted performance of the related schedules, thus the order of improvements in SIP schedules
reported here remains.
Stochastically Optimal Pit Limits. The previous comparisons were based on the same pit limits
deemed optimal using best industry practice [1]. This section focuses on the value of the proposed
approaches with respect to stochastically optimal pit limits. Both methods described above consider
larger pit limits, and stop when discounted cash flows are no longer positive. Figures 12 and 13 show
some of the results. The stochastically generated optimal pit limits contain an additional 15 % of
tonnage when compared to the traditional (deterministic) “optimal” pit limits, add about 10 % in NPV
to the NPV reported above from stochastic production scheduling within the conventionally optimal pit
limits, and extend the life-of-mine. These are substantial differences for a mine of a relatively small
size and short life-of-mine. Further work shows that there are additional improvements on all aspects
when a stochastic framework is used for mine design and production scheduling.

147
 Fig. 11. Cross-sectional views of the SIP (bottom) and traditional schedule (TS - top) for a copper deposit

Fig. 12. LOM cumulative cash flows for the conventional approach, simulated annealing and SIP, and is
compared to results from conventionally derived optimal pit limits

Fig. 13. Stochastic pit limits are larger than the conventional ones; physical scheduling differences are
expected when bigger pits are generated

The new approach yielded an increment of ~ 30 % in the NPV when compared to the conventional
approach. The differences reported are due to the different scheduling patterns, the waste mining rate,
and an extension of the pit limits, which yielded an additional ~ 5.5 thousand tonnes of metal.
CONCLUSIONS
Starting from the limits of the current orebody modeling and life-of-mine planning optimization
paradigm, an integrated risk-based framework has been presented. This framework extends the common
approaches in order to integrate both stochastic modeling of orebodies and stochastic optimization in a
complementary manner. The main drawback of estimation techniques and traditional approaches to planning
is that they are unable to account for the in-situ spatial variability of the deposit grades; in fact, conventional
148
optimizers assume perfect knowledge of the orebody being considered. Ignoring this key source of risk and
uncertainty can lead to unrealistic production expectations as well as suboptimal mine designs.
The work presented herein shows that the stochastic framework adds higher value in production
schedules in the order of 25 %, and will be achieved regardless of which method from the two
presented is used. Furthermore, stochastic optimal pit limits are shown to be about 15 % larger in terms
of total tonnage, compared to the traditional (deterministic) optimal pit limits. This difference extends
the life-of-mine and adds approximately 10 % of net present value (NPV) to the NPV reported above
from stochastic production scheduling within the conventionally optimal pit limits.
Thanks are in order to the International Association of Mathematical Geosciences for the opportunity
to present this work as their distinguished lecturer. The support of AngloGold Ashanti, Barrick, BHP
Billiton, De Beers, Newmont, Vale and Vale Inco, as well as the National Science and Engineering
Research Council (NSERC) of Canada, Canada Research Chairs Program, and Canadian Foundation for
Innovation (CFI) are gratefully acknowledged. Thanks to R. Goodfellow for editorial assistance.

REFERENCES
1. J. Whittle, “A decade of open pit mine planning and optimisation – the craft of turning algorithms into
packages,” in: APCOM'99 Computer Applications in the Minerals Industries 28th International Symposium,
Colorado School of Mines, Golden (1999).
2. M. David, Handbook of Applied Advanced Geostatistical Ore Reserve Estimation, Elsevier Science
Publishers, Amsterdam (1988).
3. R. Dimitrakopoulos, C.T. Farrelly, and M. Godoy, “Moving forward from traditional optimization: grade
uncertainty and risk effects in open-pit design,” Transactions of the Institution of Mining and Metallurgy,
Section A: Mining Technology, 111 (2002).
4. A. G. Journel, “Modelling uncertainty: some conceptual thoughts,” in: Geostatistics for the Next Century,
R. Dimitrakopoulos (Ed.), Kluwer Academic Publishers, Dordrecht (1994).
5. M. Kent, R. Peattie, and V. Chamberlain, “Incorporating grade uncertainty in the decision to expand the
main pit at the Navachab gold mine, Namibia, through the use of stochastic simulation”, The Australasian
Institute of Mining and Metallurgy, Spectrum Series, 14 (2007).
6. M. C. Godoy and R. Dimitrakopoulos, “A risk analysis based framework for strategic mine planning and
design – Method and application,” Journal of Mining Science, No. 2 (2011).
7. M. C. Godoy and R. Dimitrakopoulos, “Managing risk and waste mining in long-term production
scheduling,” SME Transactions, 316 (2004).
8. H. Mustapha and R. Dimitrakopoulos, “High-order stochastic simulations for complex non-Gaussian and
non-linear geological patterns,” Mathematical Geosciences, 42, No. 5 (2010).
9. S. A. Abdel Sabour and R. Dimitrakopoulos, “Accounting for joint ore supply, metal price and exchange
rate uncertainties in mine design,” Journal of Mining Science, No. 2 (2011).
10. C. Meagher, S. A. Abdel Sabour, and R. Dimitrakopoulos, “Pushback design of open pit mines under
geological and market uncertainties,” The Australasian Institute of Mining and Metallurgy, Spectrum Series
No. 17 (2010).
11. R. Dimitrakopoulos, H. Mustapha, and E. Gloaguen, “High-order statistics of spatial random fields:
Exploring spatial cumulants for modelling complex, non-Gaussian and non-linear phenomena,”
Mathematical Geosciences, 42, No. 1 (2010).
12. B. D. Ripley, Stochastic Simulations, J. Wiley & Sons, New York (1987).
13. A. M. Law and W. D. Kelton, Simulation Modeling and Analysis, McGraw-Hill Higher Education,
Singapore (1999).
14. R. Dimitrakopoulos and X. Luo, “Generalized sequential Gaussian simulation on group size ν and screen-
effect approximations for large field simulations,” Mathematical Geology, 36 (2004).
149
 15. H. Mustapha and R. Dimitrakopoulos “Generalized Laguerre expansions of multivariate probability densities
with moments,” Computers & Mathematics with Applications, 60, No. 7 (2010).
16. A. Leite and R. Dimitrakopoulos, “A stochastic optimization model for open pit mine planning: Application
and risk analysis at a copper deposit,” Transactions of the Institution of Mining and Metallurgy: Mining
Technology, 116, No. 3 (2007).
17. F. Albor Consquega and R. Dimitrakopoulos, “Stochastic mine design optimization based on simulated
annealing: Pit limits, production schedules, multiple orebody scenarios and sensitivity analysis,”
Transactions of the Institution of Mining and Metallurgy: Mining Technology, 118, No. 2 (2009).
18. S. Ramazan and R. Dimitrakopoulos, “Stochastic optimisation of long-term production scheduling for open
pit mines with a new integer programming formulation,” The Australasian Institute of Mining and
Metallurgy, Spectrum Series, 14 (2007).
19. S. Ramazan and R. Dimitrakopoulos, “Production scheduling with uncertain supply – A new solution to the
open pit mining,” COSMO Research Report, No. 2, McGill University, Montreal (2008).
20. M. Menabde, G. Froyland, P. Stone, and G. Yeates, “Mining schedule optimisation for conditionally
simulated orebodies,” The Australasian Institute of Mining and Metallurgy, Spectrum Series, 14 (2007).
21. A. Leite and R. Dimitrakopoulos, “Production scheduling under metal uncertainty – Application of stochastic
mathematical programming at an open pit copper mine and comparison to conventional scheduling,” The
Australasian Institute of Mining and Metallurgy, Spectrum Series, 17 (2010).
22. S. Geman and D. Geman, “Stochastic relaxation, Gibbs distribution, and the Bayesian restoration of images,”
IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-6, 6 (1984).
23. H. Lerchs and I. F. Grossmann, “Optimum design of open pit mines,” CIM Bulletin, Canadian Institute of
Mining and Metallurgy, 58 (1965).
24. R. Dimitrakopoulos and S. Ramazan, “Stochastic integer programming for optimizing long term production
schedules of open pit mines: methods, application and value of stochastic solutions,” Transactions of the
Institution of Mining and Metallurgy, Section A: Mining Technology, 117, No. 4 (2008).
25. R. Dimitrakopoulos and S. Ramazan, “Uncertainty based production scheduling in open pit mining,” SME
Transactions, 316 (2004).

150

View publication stats