Вы находитесь на странице: 1из 6

Control theory in mine planning and management:

algorithms for the open pit mine optimal scheduling


Michel De Lara
Universite Paris-Est, Cermics,

Ecole des Ponts ParisTech


6-8 avenue Blaise Pascal, 77455 Marne la Vallee Cedex 2, France.
delara@cermics.enpc.fr
September 26, 2010
Abstract
We claim that a dynamic optimal control approach may provide interesting insights to solve
the open pit mine optimal scheduling. We present some challenges and some numerical exper-
iments. We conclude on perspectives on how control theory may help to handle uncertainties
and environmental impacts in decision-making.
1 Introduction
Open pit mines are generally designed with the help of so-called block models, or resource models.
These latter represent the material inside the pit using millions of blocks with given shape. Open pit
mines extraction is submitted to physical restrictions: only blocks at the surface may be extracted; a
block cannot be extracted if the slope made with one of its neighbors is too high, due to geotechnical
constraints on mine wall slopes.
In essence, exploiting an open pit mine can be seen as selecting an admissible extraction sequence
of blocks where, by admissible, we mean that the above physical restrictions are satised. Among
all admissible extraction sequences, of particular economic interest are those which are the more
protable.
To be more specic, suppose it takes one time unit to extract one block, and that a retirement
option is available where no block is extracted. Assume that the block model is completed with a
prot model giving the economic value of each block (benet minus extraction costs), according to
grade, costs and metal prices depending on market constraints. We shall consider the case where
this prot model is deterministic, though the methods we present may be extended to the uncertain
case. Then, given a discount factor per time unit, the net present value (NPV) of an admissible
extraction sequence is the discounted sum of extracted block economic values. Characterizing
admissible extraction sequences which maximize the net present value is refered as open pit mine
sequencing optimization, or open pit mine optimal scheduling.
Such problem of NPV optimized extraction sequence is much less addressed than the so-called
optimal nal pit problem, or optimization pit limit design, which characterize the nal conguration
of the open pit at the end of its economic life. Lerchs and Grossman [3] demonstrated that the
1
optimal pit limit problem is equivalent to determining the maximal closure in a directed, weighted
graph, where geotechnical constraints on mine wall slopes are included as a set of precedence
constraints.
Most of the techniques in mine planning are from Operations Research [4]. Typical algorithms
used formulate the problem as a mixed integer linear programme. Network ow algorithms and
so-called oating cone techniques have, among others, completed the panoply of pit shell design
algorithms.
2 An optimal control insight into mine planning
The Lerchs and Grossman approach is static, in that it focuses on the nal pit and assumes
the discount factor to be one (i. e. the discount rate to be zero). Our focus is on open pit
mine sequencing optimization (which includes optimization pit limit design) and on estimating the
optimal NPV.
To test sensitivity of the optimal NPV to costs and metal prices, one should ideally dispose of
fast algorithms to compute an optimal scheduling. However, due to the high number of blocks,
such issue is out of reach. We claim that formulating the open pit mine sequencing optimization
problem as one of optimal control as in [2] (see also [1]) provides optimal algorithms and can suggest
suboptimal ones.
Theoretically, the open pit mine sequencing optimization can be solved by dynamic program-
ming. However, its practical application suers from the curse of the dimensionality. We briey
sketch some ideas and challenges to bypass this issue, and we suggest extensions.
2.1 Exact algorithms for open pit mine optimal scheduling
For the moment, we have developped a computer code which solves in a few minutes the
dynamic programming equation for 2-dimensional (2D) mines with 11 columns each having
5 blocks [1]. Suppose that each of these 55 blocks is in fact a super-block corresponding to
the material which can be extracted in one year. The economic value of each super-block is the
(non discounted) sum of the economic value of all original blocks within. With this agregated
approach, we are able to provide the open pit mine optimal scheduling over 36 years
1
. By
improving the mathematical coding of admissible mine proles, we expect to successfully
attack slightly larger mines.
We still have to write a computer code which solves the dynamic programming equation for
3-dimensional (3D) mines. With our current techniques, a cubic mine with 4 4 4 blocks
can be treated, which corresponds to 20 years with the yearly super-block approach above.
By improving the mathematical coding of admissible mine proles, we expect to successfully
attack 5 5 5 blocks, corresponding to 35 years, and maybe slightly larger mines and other
geometries.
These exact algorithms remain valid when the prot model is random, with i.i.d. prices. Markov
prices models increase the dimension by one.
1
36=11+9+7+3+1 super-blocks admissible for extraction at most.
2
2.2 Suboptimal algorithms for open pit mine optimal scheduling
For the moment, we have developped a computer code which attacks very large 2D and 3D
mines (several million blocks) in a few minutes by an index suboptimal algorithm. Results
depend on the prot model, and many more models should be tested than the few ones we
have.
We have ideas of other index suboptimal algorithms, which have the quality to be rather
insensitive to the curse of the dimensionality, hence can attack fastly realistic mines models.
We need to develop the corresponding computer codes.
These suboptimal algorithms can be adapted when the prot model is random.
2.3 Numerical examples in 2D
We present 2D examples, with an annual interest rate of r
f
=0% or r
f
=10%, and we provide
an upper bound for the NPV by ranking all blocks by decreasing value, and by taking the
discounted sum;
the NPV given by the greedy policy which consists in selecting at each period the admissible
top block with highest economic value;
the NPV given by the best of a family of index policies;
when possible, i. e. for a small 2D mine, the optimal NPV.
2.3.1 Comparison between optimal DP and suboptimal policies for a small size mine
For a small size 2D mine (11 columns each having 5 blocks), we can compare index policies with the
optimal policy given by dynamic programming (DP). We suppose that each block is a super-block
corresponding to the material which can be extracted in one year. Hence, the discount factor equals
=
1
1+r
f
.
We consider a portion of the Lerchs and Grossman mine example in [3] with C = 11 columns
and D = 5 blocks in each. The prot model is given by
- 4. 8. 12. 12. 0. - 4. - 4. - 4. - 4. - 4. - 4.
- 4. 0. 12. 12. 8. - 4. - 4. - 4. - 4. - 4. - 4.
- 4. - 4. 8. 12. 12. 0. - 4. - 4. - 4. - 4. - 4.
- 4. - 4. 0. 12. 12. 8. - 4. - 4. - 4. - 4. - 4.
- 4. - 4. - 4. 8. 12. 12. 0. - 4. - 4. - 4. - 4.
The optimal NPV is computed in a few minutes and suboptimal index policies give a NPV in
a few seconds. Comparisons are presented in Table 1, and proles are shown in Figure 1.
2.3.2 Comparison between suboptimal policies for a large size mine
For a larger size 3D mine, the optimal policy given by DP is out of numerical reach. We now
consider a vertical cut of the Marvin mine example. We suppose that 1 500 blocks are extracted
in one year, hence the discount factor is such that
1 500
=
1
1+r
f
.
The suboptimal index policies are computed in a few seconds. Comparisons are presented in
Table 2, and proles are shown in Figure 2.
3
Policy /bound upper bound optimal best index greedy
NPV for r
f
=0% 160 104 72 56
NPV for r
f
=10% 93.97 53.57 47.53 41.81
Table 1: Comparison of NPV given by three extraction strategies for a portion of the Lerchs and
Grossman mine example
0 2 4 6 8 10 12
5
4
3
2
1
0
Optimal extraction profile for annual interest rate 10%
(a) Optimal r
f
= 10%
0 2 4 6 8 10 12
5
4
3
2
1
0
Best index policy final pit profile for annual interest rate 10%
(b) Best index r
f
= 10%
0 2 4 6 8 10 12
5
4
3
2
1
0
Greedy policy final pit profile for annual interest rate 10%
(c) Greedy r
f
= 10%
0 2 4 6 8 10 12
5
4
3
2
1
0
Optimal extraction profile for annual interest rate 0%
(d) Optimal r
f
= 0%
0 2 4 6 8 10 12
5
4
3
2
1
0
Best index policy final pit profile for annual interest rate 0%
(e) Best index r
f
= 0%
0 2 4 6 8 10 12
5
4
3
2
1
0
Greedy policy final pit profile for annual interest rate 0%
(f) Greedy r
f
= 0%
Figure 1: Comparison of nal pits for a portion of the Lerchs and Grossman mine example for
annual interest rates r
f
= 10% and r
f
= 0% (the darker, the richer)
Policy/bound upper bound optimal best index greedy
NPV for r
f
= 0% 142.73 ? 75.86 41.15
NPV for r
f
= 10% 141.92 ? 75.53 41.04
Table 2: Comparison of NPV given by two extraction strategies for a vertical cut of the Marvin
mine example
4
0 10 20 30 40 50 60
16
14
12
10
8
6
4
2
0
Greedy policy final pit profile for annual interest rate 0%
(a) Greedy nal r
f
= 0%
0 10 20 30 40 50 60
16
14
12
10
8
6
4
2
0
Best index policy final pit profile for annual interest rate 0%
(b) Index nal r
f
= 0%
0 10 20 30 40 50 60
16
14
12
10
8
6
4
2
0
Greedy policy final pit profile for annual interest rate 10%
(c) Greedy nal r
f
= 10%
0 10 20 30 40 50 60
16
14
12
10
8
6
4
2
0
Best index policy final pit profile for annual interest rate 10%
(d) Index nal r
f
= 10%
Figure 2: Comparison of suboptimal strategies for a vertical cut of the Marvin mine example for
annual interest rates r
f
= 0% and r
f
= 10% (the darker, the richer)
5
3 Perspectives
We have presented preliminary results on how a dynamic optimal control approach may provide
interesting insights to solve the open pit mine optimal scheduling. Such an approach may be
extended to the uncertain case, providing ways to take into account risk and information in mine
planning (see [1]).
The challenges raised by the environmental impacts of mining contribute to looking at other
criteria than the only NPV, such as pollution, and at focusing on constraints due to ecological or
health thresholds. The control theory approach may again be of some help in allowing to integrate
the long time horizon of mine planning, the uncertainties, and a priori conicting ecological, health
and economic constraints. We expect to successfully apply to mine planning under environmental
constraints the methods to be found in [1], which have already been tried in the management of
other natural resources.
References
[1] M. De Lara and L. Doyen. Sustainable Management of Natural Resources. Mathematical Models
and Methods. Springer-Verlag, Berlin, 2008.
[2] G.C. Goodwin, M.M. Seron, R.H. Middleton, M. Zhang, B.F. Hennessy, P.M. Stone, and
M. Menabde. Receding horizon control applied to optimal mine planning. Automatica,
42(8):13371342, 2006.
[3] H. Lerchs and I. F. Grossman. Optimum design of open-pit mines. Transactions, CIM, 68:1724,
1965.
[4] Alexandra M. Newman, Enrique Rubio, Rodrigo Caro, Andres Weintraub, and Kelly Eurek. A
Review of Operations Research in Mine Planning. INTERFACES, 40(3):222245, 2010.
6

Вам также может понравиться