Sei sulla pagina 1di 18

A MIP model for underground mine planning

Fr ed eric Meunier Work in progress with Nelson Morales

Mine planning is a core discipline in mining. It starts with the resources and generates a production plan (a bankable document). Mine planning is hard, therefore it has been split into several steps. Dening what to mine (distinguishing reserves from resources) is decoupled from determining how to do it (not only conceptually, but often done by dierent teams/consultants).

Underground mining: two kinds of tasks Preparation (construction of tunnels, shafts, crosscuts) Extraction Preemption is allowed. Due to physical constraints, these various tasks are linked by precedence constraints (typically: 3000 tasks) To perform these tasks, ressources are necessary. Ressources available but limited. Each percent of task i, when achieved, leads to a prot (extaction task) or a cost (preparation task) that depends on time.

Scheduling the tasks: a matter of prot There is a need for a tool providing a schedule of the tasks that maximizes the prot. This schedule has to be at a strategic level: time is discretized in periods t = 1, . . . , T. A period (typically: about 30 days). (Typical number of periods: 55). For each period t: the prot of performing a unit of task i: vit . the availability of resource r: Kt r.

Notations V: set of tasks A: precedence constraints between tasks T periods (t = 1, . . . , T) (period t takes Dt days) R: set of resource types (for instance: tunnel boring machines, crushers, water pumps, workers...) Kt r : set of resources of type r available during period t 1% of task i needs cri days of use of a type r resource (parallelization is allowed).

A rst MIP formulation variables st i : equal to 1 if task i has started by period t and 0 if not variables et i : equal to 1 if task i has not yet ended by period t and 0 if not variables pt i : percentage of task i developed during the period t

A rst MIP formulation max s.t. vit pt i st+1 i 1 et i st i et i i for all i for all i for all i for all i for all i for all i
for all for all

i,t

st i et i pt i pt i 1 et i t st i , ei pt i

ps i {0, 1} 0
s t

V, t V, t V, t V, t V, t V, t V, t

t1 st j 1 ei t i cri pi t |Kt r |D

(i, j) A, t r R, t

for all

This model has a drawback The constraint


t1 st for all (i, j) A, t = 2, . . . , T j 1 ei

prohibits feasible schedule: if task i ends at the beginning of period t, the remaining time could be used to start task j.

A less strict constraint


t st j 1 ei for all (i, j) A, t = 1, . . . , T

may allow unfeasible schedule: task i may nish at the end of period t and task j at its beginning.

Morales, Rocher and Rubio (2011) proposed a way to partially settle the problem and succeeded in allowing more feasible solutions. However, their model needs an exponential number of constraints and some feasible schedules are still not covered.

The one period problem

V: set of tasks A: precedence constraints between tasks D days available R: set of ressource types (for instance: tunnel boring machines, crushers, water pumps, workers...) Kr : set of resources of type r task i needs cri days of use of a type r ressource (parallelization is allowed). Is it possible to achieve all tasks in D days ?

The one period problem: necessary condition Assume that we have a feasible schedule and denote: i the number of days devoted to task i dri the cumulated number of days a resource of type r is assigned to task i i the number of days the resource is assigned to task i Then we have i D for all resources
iV

dri =
Kr

i for all tasks i and resource types r i


rR,Kr

max

i for all tasks i

i D for any path P in the digraph (V, A)


iP

dri = cri for all tasks i and resource types r

The one period problem: sucient condition However, it is not a sucient condition (counter-examples exist). Nevertheless if i D for all resources
iV

dri =
Kr

i for all tasks i and resource types r i = max dri for all tasks i
rR

i D for any path P in the digraph (V, A)


iP

dri = cri for all tasks i and resource types r then there exists a feasible schedule.

iP i D for any path P in the digraph (V, A) requires an exponential description. Introducing variables yi for i V reduces the number of inequalities: yi = 0 if i has no predecessor yi D if i has no successor yj yi + i for all (i, j) A

The one period problem: sucient condition compact version Proposition If there exist yi for all i V such that cri |Kr |D for each resource type r yi = 0 if i has no predecessor yi D if i has no successor yj yi + maxrR cri for all (i, j) A
iV

then there is a feasible scheduling. This condition can be tested through linear programming. More eciently, through dynamic programming: once iV cri |Kr |D for each resource type r we can use a PERT approach: put on each arc (i, j) a weight = maxrR cri ; feasible schedule: i the longest path in the graph has length D.

Back to the multi-period problem

The constraints are added: t t iV cri |Kr |D for each resource type r. yi = 0 if i has no predecessor yi D if i has no successor yj yi + maxrR ct ri for all (i, j) A

Back to the multi-period problem: an advanced MIP formulation We get a new model (but still with feasible solutions not covered) max s.t.
i,t

vit pt i usual constraints dening t t st , i ei and si st j yi yi yj = 1 et i 0 Dt yi + maxrR ct ri


t |Kt r |D
for all

(i, j) A, t

if i has no predecessor if i has no successor for all (i, j) A

t i cri pi

for all

r R, t

Open questions

For the one-period problem, is it possible to characterize feasible instances ? Is there even a polynomial algorithm ? A positive answer would lead to a compact MIP formulation of the multi-period problem.

Potrebbero piacerti anche