Introduction

Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Stochastic Discrete Multiple Objective

30th International Conference of Jangjeon Mathematical Society

Pur and Applied Mathematics

ICJMS USTHB-ALGERIA 2017

Operations Research Department

Faculty of Mathematics - USTHB - ALGERIA

July 19, 2017

Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

A N ew A lgorithme for Determinating the whole S et of I nteger E fficient S tochastic S olutioins

I T he problem consists of searching for the solution

Pareto-set in stochastic environment, when, p criteria are
to be optimized simultaneously.
 Consider (Ω, Ξ, P) a probability space, with discrete
distribution.
 Given a Multiple Objective Integer Stochastic Linear
Programming problem (MOISLP)


 “ min ”Fk = ck (ξ)x; k = 1, · · · , p
s.t. Ax = b;

(PSC )

 T(ξ)x = h(ξ);
x∈N


• A, b matrices: (m × n), (m × 1): decision constraint

• ck , T, h are random matrices : (K × n), (` × n) and (` × 1)
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

M otivations : Why optimizing over integer efficient set?

The problem is well known when the parame-

ters are deterministic; unfortunately, it is rare in
stochastic environment/
As far as we know, the only method that solves
such problem is published in (EJOR 2006) .
This does not provide all integer efficient solu-
tions.

Some real world problems have to be modeled

as (PSC ).

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

M otivations : Why optimizing over integer efficient set?

The problem is well known when the parame-

ters are deterministic; unfortunately, it is rare in
stochastic environment/
As far as we know, the only method that solves
such problem is published in (EJOR 2006) .
This does not provide all integer efficient solu-
tions.

Some real world problems have to be modeled

as (PSC ).

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

M otivations : Why optimizing over integer efficient set?

The problem is well known when the parame-

ters are deterministic; unfortunately, it is rare in
stochastic environment/
As far as we know, the only method that solves
such problem is published in (EJOR 2006) .
This does not provide all integer efficient solu-
tions.

Some real world problems have to be modeled

as (PSC ).

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

M otivations : Why optimizing over integer efficient set?

The problem is well known when the parame-

ters are deterministic; unfortunately, it is rare in
stochastic environment/
As far as we know, the only method that solves
such problem is published in (EJOR 2006) .
This does not provide all integer efficient solu-
tions.

Some real world problems have to be modeled

as (PSC ).

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

H ypothesis
ä We suppose some probability space (Ω, Ξ, P) that defines
the probabilistic aspect of random parameters in our
problem.

ä The decision on x has to be taken before the realization of

the random variables is known “here and now".

ä We suppose a fixed recourse matrix W and the recourse

costs is taken as linear defined by q0 y

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

H ypothesis
ä We suppose some probability space (Ω, Ξ, P) that defines
the probabilistic aspect of random parameters in our
problem.

ä The decision on x has to be taken before the realization of

the random variables is known “here and now".

ä We suppose a fixed recourse matrix W and the recourse

costs is taken as linear defined by q0 y

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

H ypothesis
ä We suppose some probability space (Ω, Ξ, P) that defines
the probabilistic aspect of random parameters in our
problem.

ä The decision on x has to be taken before the realization of

the random variables is known “here and now".

ä We suppose a fixed recourse matrix W and the recourse

costs is taken as linear defined by q0 y

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

ä Given the penalties qr = q(ξ r ) to the constraint violations

are given
R
X R
X
ä F
ek (x) = E(Fk (x)) = pr Fk,r (x) = pr ck (ξ r )x
r=1 r=1

ä A recourse function Q(x, ξ r ) is added to each criterion Fkr ,

ä the corresponding penalty is given by

Q(x, ξ r ) = min {(qr )t f |W(ξ r )f = h(ξ r ) − T(ξ r )x; f ≥ 0}

f

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

ä Given the penalties qr = q(ξ r ) to the constraint violations

are given
R
X R
X
ä F
ek (x) = E(Fk (x)) = pr Fk,r (x) = pr ck (ξ r )x
r=1 r=1

ä A recourse function Q(x, ξ r ) is added to each criterion Fkr ,

ä the corresponding penalty is given by

Q(x, ξ r ) = min {(qr )t f |W(ξ r )f = h(ξ r ) − T(ξ r )x; f ≥ 0}

f

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

ä Given the penalties qr = q(ξ r ) to the constraint violations

are given
R
X R
X
ä F
ek (x) = E(Fk (x)) = pr Fk,r (x) = pr ck (ξ r )x
r=1 r=1

ä A recourse function Q(x, ξ r ) is added to each criterion Fkr ,

ä the corresponding penalty is given by

Q(x, ξ r ) = min {(qr )t f |W(ξ r )f = h(ξ r ) − T(ξ r )x; f ≥ 0}

f

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

ä Given the penalties qr = q(ξ r ) to the constraint violations

are given
R
X R
X
ä F
ek (x) = E(Fk (x)) = pr Fk,r (x) = pr ck (ξ r )x
r=1 r=1

ä A recourse function Q(x, ξ r ) is added to each criterion Fkr ,

ä the corresponding penalty is given by

Q(x, ξ r ) = min {(qr )t f |W(ξ r )f = h(ξ r ) − T(ξ r )x; f ≥ 0}

f

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

ä The deterministic multiple objective integer linear

programming problem

 R
X
pr crk x

“ min ” F = (F ) = k = 1, · · · , p


   e k E k
P
e r=1


 s.t. Ax = b
x ∈ Nn


Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

• The parametric problem is given by

p

  min (E(F)) = X λ E(F )

λ k k
P
eλ
 k=1
s.t. x ∈ D = {x ∈ Rn |Ax = b, x ∈ N}


λ = (λ)k=1,··· ,p : is an integer parameter vector.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

• The parametric problem is given by

p

  min (E(F)) = X λ E(F )

λ k k
P
eλ
 k=1
s.t. x ∈ D = {x ∈ Rn |Ax = b, x ∈ N}


λ = (λ)k=1,··· ,p : is an integer parameter vector.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Deterministic E quivalent Problem

ä The parametric deterministic problem including penalty

variable

p

  min (E(F)) = X λ E(F ) + θ
λ k k
P
eλ
 k=1
s.t. x∈D


Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
The Problem
Principal Components of the Technique
Motivations
The algorithm
Passage to deterministic equivalent problem of MOSILP
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

ä First Stage
Feasibility and optimality tests as was introduced in
L-shapped technique First Set

ä Second Stage
Basic definitions and results in multiple objective in-
teger linear programming theory
Second Set

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

ä First Stage
Feasibility and optimality tests as was introduced in
L-shapped technique First Set

ä Second Stage
Basic definitions and results in multiple objective in-
teger linear programming theory
Second Set

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

A dmissibility T est

• To check out for feasibility of the second stage-problems,

we solve the following problem :

For all possible realizations of ξ

0 h(ξ r ) − T(ξ r )x0




 max u
s.t. ut W ≤ 0

(FDual )

 kuk1 ≤ 1
u ∈ R`


Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

A dmissibility T est

• In case where utr h (ξ r ) − T (ξ r ) x0 > 0 for some

 

ξ r ; r ∈ {1, · · · , R} and ur is an optimal solution of the

previous problem ;
• +
Admissubility cut utr [h (ξ r ) − T (ξ r ) x] ≤ 0

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

A dmissibility T est

• In case where utr h (ξ r ) − T (ξ r ) x0 > 0 for some

 

ξ r ; r ∈ {1, · · · , R} and ur is an optimal solution of the

previous problem ;
• +
Admissubility cut utr [h (ξ r ) − T (ξ r ) x] ≤ 0

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

A dmissibility T est

• To test the optimality of a given solution, we solve the

following problem :

For all possible realizations of ξ

 max π 0 (h(ξ r ) − T(ξ r )x)



(Ftest ) s.t. π t W ≤ (qr )0

π ∈ R`


Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

A dmissibility T est

R
X
• Q(x0 ) = p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1

• if Q(x0 ) ≤ θ then x0 is optimal solution

• else + optimality cut

R
X
θ≥ pi πit (hi − Ti x)
i=1
Return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

A dmissibility T est

R
X
• Q(x0 ) = p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1

• if Q(x0 ) ≤ θ then x0 is optimal solution

• else + optimality cut

R
X
θ≥ pi πit (hi − Ti x)
i=1
Return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

A dmissibility T est

R
X
• Q(x0 ) = p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1

• if Q(x0 ) ≤ θ then x0 is optimal solution

• else + optimality cut

R
X
θ≥ pi πit (hi − Ti x)
i=1
Return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of the Domain

ä At iteration `, using Sylva and Crema’s idea (EJOR 2007)

ä The feasible set D is reduced by

eliminating all dominated solutions by Cx̂`

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of the Domain

ä At iteration `, using Sylva and Crema’s idea (EJOR 2007)

ä The feasible set D is reduced by

eliminating all dominated solutions by Cx̂`

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of the Domain

ä At iteration `, using Sylva and Crema’s idea (EJOR 2007)

ä The feasible set D is reduced by

eliminating all dominated solutions by Cx̂`

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

let x1 , x2 , ..., x` be efficient solutions to problem (P) and

∆k = x ∈ Fn | ecx ≥ ecxk (k ∈ 1, · · · , `).
if x? is an optimal solution to
`
( !)
G
` 0
(Pλ ) = min λ ecx | x ∈ D e− ∆k
k=1

for some λ ∈ Rp ,λ > 0, then x? is an efficient solution to

problem (PSc )

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

T he main Techniques

ä Resolution of the parametric problem (P

eλ )

ä Two Tests for the obtained solution x∗

• Feasibility • Optimality

ä Updating the list of efficient solutions

ä Reducing the admissible region

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

T he main Techniques

ä Resolution of the parametric problem (P

eλ )

ä Two Tests for the obtained solution x∗

• Feasibility • Optimality

ä Updating the list of efficient solutions

ä Reducing the admissible region

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

T he main Techniques

ä Resolution of the parametric problem (P

eλ )

ä Two Tests for the obtained solution x∗

• Feasibility • Optimality

ä Updating the list of efficient solutions

ä Reducing the admissible region

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

T he main Techniques

ä Resolution of the parametric problem (P

eλ )

ä Two Tests for the obtained solution x∗

• Feasibility • Optimality

ä Updating the list of efficient solutions

ä Reducing the admissible region

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

T he main Techniques

ä Resolution of the parametric problem (P

eλ )

ä Two Tests for the obtained solution x∗

• Feasibility • Optimality

ä Updating the list of efficient solutions

ä Reducing the admissible region

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

T he main Techniques

ä Resolution of the parametric problem (P

eλ )

ä Two Tests for the obtained solution x∗

• Feasibility • Optimality

ä Updating the list of efficient solutions

ä Reducing the admissible region

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

The Algorithm
Initial Step ` ← 1 , End ← false , L` = ∅ , θ ← −∞
e` ← D
et D e
The parametric problem

p

  min E(F ) = X λ E(F )
λ k k
P
eλ
 k=1
s.t. x∈ D


where: D = {x ∈ Rn |Ax ≤ b, x ∈ N} is being solved.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

The Algorithm
Initial Step ` ← 1 , End ← false , L` = ∅ , θ ← −∞
e` ← D
et D e
The parametric problem

p

  min E(F ) = X λ E(F )
λ k k
P
eλ
 k=1
s.t. x∈ D


where: D = {x ∈ Rn |Ax ≤ b, x ∈ N} is being solved.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

• if (P
eλ ) is not feasible then (P(SC) ) is not feasible

END

• else, x0 is an optimal solution

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

• if (P
eλ ) is not feasible then (P(SC) ) is not feasible

END

• else, x0 is an optimal solution

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

• if (P
eλ ) is not feasible then (P(SC) ) is not feasible

END

• else, x0 is an optimal solution

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario Example

The problem (FDual ) is being solved. Test of Feasibility

β ← ut h (ξ r ) − T (ξ r ) x`
 

• If β > 0
Update the current admissible region by

utr [h (ξ r ) − T (ξ r ) x] ≤ 0

• If β = 0 for all scenarios, x0 is feasible

• Test_Optimality

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario

• We solve the problem (Ftest ) . Ftest

R
X
• Q(x0 ) ← p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1
• If Q(x0 ) > θ

+ optimality cut

R
X
p(ξ r ) [(hξ r ) − T(ξ r )x] ≤ θ
r=1

• We solve the problem . Fopt

• else Q(x0 ) = θ
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario

• We solve the problem (Ftest ) . Ftest

R
X
• Q(x0 ) ← p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1
• If Q(x0 ) > θ

+ optimality cut

R
X
p(ξ r ) [(hξ r ) − T(ξ r )x] ≤ θ
r=1

• We solve the problem . Fopt

• else Q(x0 ) = θ
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario

• We solve the problem (Ftest ) . Ftest

R
X
• Q(x0 ) ← p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1
• If Q(x0 ) > θ

+ optimality cut

R
X
p(ξ r ) [(hξ r ) − T(ξ r )x] ≤ θ
r=1

• We solve the problem . Fopt

• else Q(x0 ) = θ
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario

• We solve the problem (Ftest ) . Ftest

R
X
• Q(x0 ) ← p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1
• If Q(x0 ) > θ

+ optimality cut

R
X
p(ξ r ) [(hξ r ) − T(ξ r )x] ≤ θ
r=1

• We solve the problem . Fopt

• else Q(x0 ) = θ
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario

• We solve the problem (Ftest ) . Ftest

R
X
• Q(x0 ) ← p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1
• If Q(x0 ) > θ

+ optimality cut

R
X
p(ξ r ) [(hξ r ) − T(ξ r )x] ≤ θ
r=1

• We solve the problem . Fopt

• else Q(x0 ) = θ
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario

• We solve the problem (Ftest ) . Ftest

R
X
• Q(x0 ) ← p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1
• If Q(x0 ) > θ

+ optimality cut

R
X
p(ξ r ) [(hξ r ) − T(ξ r )x] ≤ θ
r=1

• We solve the problem . Fopt

• else Q(x0 ) = θ
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For each Scenario

• We solve the problem (Ftest ) . Ftest

R
X
• Q(x0 ) ← p(ξ r )(πr )t [h(ξ r ) − T(ξ r )x0 ]
r=1
• If Q(x0 ) > θ

+ optimality cut

R
X
p(ξ r ) [(hξ r ) − T(ξ r )x] ≤ θ
r=1

• We solve the problem . Fopt

• else Q(x0 ) = θ
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Update

(x` , θ` ) ← (e
x` , θe` ) efficient stochastic solution .
• ` `
Fi = Fi (x ), for i = 1, · · · , p non dominated criterion
e e
vector

• List of efficient solutions: L` = L`−1 ∪ {F

e`}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Update

(x` , θ` ) ← (e
x` , θe` ) efficient stochastic solution .
• ` `
Fi = Fi (x ), for i = 1, · · · , p non dominated criterion
e e
vector

• List of efficient solutions: L` = L`−1 ∪ {F

e`}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of Admissible Region ` ← `+1

ä Solve the problem

p

  min E(F ) = X λ E(F )+ θ

e`λ λ k k
P
 i=1
s.t. x ∈ Dadm`


= D0

Dadm0
Dadm` = Dom`−1 ∩ ∆`

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of Admissible Region ` ← `+1

ä Solve the problem

p

  min E(F ) = X λ E(F )+ θ

e`λ λ k k
P
 i=1
s.t. x ∈ Dadm`


= D0

Dadm0
Dadm` = Dom`−1 ∩ ∆`

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of Admissible Region ` ← `+1

ä Solve the problem

p

  min E(F ) = X λ E(F )+ θ

e`λ λ k k
P
 i=1
s.t. x ∈ Dadm`


= D0

Dadm0
Dadm` = Dom`−1 ∩ ∆`

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

 

 x ∈ D eck x ≤ (eck x` − 1)y`k + Mk (1 − y`k ) 

∀k ∈ {1, · · · , p}

 

∆` = Xp
y`k ≥ 1, y`k ∈ {0, 1} ∀k ∈ {1, · · · , p}

 


 

k=1

• Mk is an upper bound to the kth objective;

• y`k ∈ {0, 1} is a binary variable associated to the criterion

Fk = ck x;

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

 

 x ∈ D eck x ≤ (eck x` − 1)y`k + Mk (1 − y`k ) 

∀k ∈ {1, · · · , p}

 

∆` = Xp
y`k ≥ 1, y`k ∈ {0, 1} ∀k ∈ {1, · · · , p}

 


 

k=1

• Mk is an upper bound to the kth objective;

• y`k ∈ {0, 1} is a binary variable associated to the criterion

Fk = ck x;

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

e` ) is not feasible then

If (P END
λ

• L` = L`−1 ∪ {F
e ` } is the list of stochastic efficient
solutions
Otherwise

Let x` be solution an optimal solution of (P`λ )

return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

e` ) is not feasible then

If (P END
λ

• L` = L`−1 ∪ {F
e ` } is the list of stochastic efficient
solutions
Otherwise

Let x` be solution an optimal solution of (P`λ )

return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

e` ) is not feasible then

If (P END
λ

• L` = L`−1 ∪ {F
e ` } is the list of stochastic efficient
solutions
Otherwise

Let x` be solution an optimal solution of (P`λ )

return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

e` ) is not feasible then

If (P END
λ

• L` = L`−1 ∪ {F
e ` } is the list of stochastic efficient
solutions
Otherwise

Let x` be solution an optimal solution of (P`λ )

return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

e` ) is not feasible then

If (P END
λ

• L` = L`−1 ∪ {F
e ` } is the list of stochastic efficient
solutions
Otherwise

Let x` be solution an optimal solution of (P`λ )

return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

e` ) is not feasible then

If (P END
λ

• L` = L`−1 ∪ {F
e ` } is the list of stochastic efficient
solutions
Otherwise

Let x` be solution an optimal solution of (P`λ )

return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Next Step

e` ) is not feasible then

If (P END
λ

• L` = L`−1 ∪ {F
e ` } is the list of stochastic efficient
solutions
Otherwise

Let x` be solution an optimal solution of (P`λ )

return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

E xample

Scenario 1 Scenario 2

“ min ”F1 = −9x1 + 4x2 “ min ”F1 = 3x1 − 2x2

 

 

“ min ”F2 = 3x1 − 5x2
 





 “ min ”F2 = 7x1 + x2

 

“ min ”F3 = 8x1 − 11x2 “ min ”F3 = −4x1 + 9x2

 


 

(Psc1 ) 
 4x1 − 2x2 ≤8 (Psc2 ) 
 4x1 − 2x2 ≤8
≤5 ≤5

  x1 + x2
 
  x1 + x2


  
 
x1 + 2x2 ≤3 ≤6
 


 s.t. 

 s.t. x1
 −2x1 + x2 ≤5  3x1 + 4x2 ≤1

 
 
 


  
 
x1 , x2 ∈N x1 , x2 ∈N

0 1 0 1
q1 = (5 3 2 1) ; P1 = q2 = (1 0 6 2) ; P2 =
2 2

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

The Deterministic problem


 “ min ”E(F1 ) = −3x1 + x2
  
“ min ”E(F2 ) = 5x1 − 2x2
P
e

 “ min ”E(F 3 ) = 2x1 − x2
s.t. x ∈ D


Où: D = {x ∈ Rn |4x1 − 2x2 ≤ 8, x1 + x2 ≤ 5, x1 , x2 ∈ N}

The parametric problem:

3
X
E(Fλ ) = λi E(Fi ) = λ1 E(F1 ) + λ2 E(F2 ) + λ3 E(F3 )
λ=1
λ1 = 6, λ2 = 1, λ3 = 1
   min E(F ) = −11x + 3x2
λ 1
P
eλ
s.t. x ∈ D
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

The Deterministic problem


 “ min ”E(F1 ) = −3x1 + x2
  
“ min ”E(F2 ) = 5x1 − 2x2
P
e

 “ min ”E(F 3 ) = 2x1 − x2
s.t. x ∈ D


Où: D = {x ∈ Rn |4x1 − 2x2 ≤ 8, x1 + x2 ≤ 5, x1 , x2 ∈ N}

The parametric problem:

3
X
E(Fλ ) = λi E(Fi ) = λ1 E(F1 ) + λ2 E(F2 ) + λ3 E(F3 )
λ=1
λ1 = 6, λ2 = 1, λ3 = 1
   min E(F ) = −11x + 3x2
λ 1
P
eλ
s.t. x ∈ D
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Step 1:Initialization parameters M1 = 5, M2 = 11, M3 = 4,

θ = −∞, D0 = D,
I Solving the weighted problem gives the solution

I eλ ) is x0 = (3 2)0
The optimal solution (P
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Step 1:Initialization parameters M1 = 5, M2 = 11, M3 = 4,

θ = −∞, D0 = D,
I Solving the weighted problem gives the solution

I eλ ) is x0 = (3 2)0
The optimal solution (P
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Admissibility Test

−4σ11 9σ21


 max +
−2σ11 3σ21 ≤


 s.t. + 0 ! !
2
σ11

−σ11 2σ21


 + ≤ 0 Max: = 3
σ21 1

(Pσ1 ) 2σ11 − 5σ21 ≤ 0 3
σ11 6σ21



 − ≤ 0
σ11 σ21

+ ≤ 1




σ11 σ21

, ∈ R

3σ12 16σ22


 max −
−2σ12 3σ22 ≤


 s.t. + 0 !
σ12
  

 −σ12 + 2σ22 ≤ 0 0
 Max: =
σ22

0
(Pσ2 ) 2σ12 − 5σ22 ≤ 0
σ12 6σ22



 − ≤ 0
σ12 σ22

+ ≤ 1




σ12 σ22

, ∈ R

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Admissibility Test

−4σ11 9σ21


 max +
−2σ11 3σ21 ≤


 s.t. + 0 ! !
2
σ11

−σ11 2σ21


 + ≤ 0 Max: = 3
σ21 1

(Pσ1 ) 2σ11 − 5σ21 ≤ 0 3
σ11 6σ21



 − ≤ 0
σ11 σ21

+ ≤ 1




σ11 σ21

, ∈ R

3σ12 16σ22


 max −
−2σ12 3σ22 ≤


 s.t. + 0 !
σ12
  

 −σ12 + 2σ22 ≤ 0 0
 Max: =
σ22

0
(Pσ2 ) 2σ12 − 5σ22 ≤ 0
σ12 6σ22



 − ≤ 0
σ12 σ22

+ ≤ 1




σ12 σ22

, ∈ R

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

•
σ1t h1 − T 1 x0 = 13 , x0 is not admissible for the first scenario, then
 

we add the admissibility cut.

•
    
2 1

1 2 x1 2 1

3 11
≥ = ⇐⇒ 5x2 ≥ 11;
3 3 −2 1 x2 3 3 5 3

•
The admissibility domain is: D(01) = D0 ∩ {x ∈ Nn | 5x2 ≥ 11}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

•
σ1t h1 − T 1 x0 = 13 , x0 is not admissible for the first scenario, then
 

we add the admissibility cut.

•
    
2 1

1 2 x1 2 1

3 11
≥ = ⇐⇒ 5x2 ≥ 11;
3 3 −2 1 x2 3 3 5 3

•
The admissibility domain is: D(01) = D0 ∩ {x ∈ Nn | 5x2 ≥ 11}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

•
σ1t h1 − T 1 x0 = 13 , x0 is not admissible for the first scenario, then
 

we add the admissibility cut.

•
    
2 1

1 2 x1 2 1

3 11
≥ = ⇐⇒ 5x2 ≥ 11;
3 3 −2 1 x2 3 3 5 3

•
The admissibility domain is: D(01) = D0 ∩ {x ∈ Nn | 5x2 ≥ 11}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

   min E(F ) = −11x + 3x2

e01 λ 1
Pλ s.t. x ∈ D01

• Solution x01 = (2 3)0 .

• Admissible for both scenarios.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

   min E(F ) = −11x + 3x2

e01 λ 1
Pλ s.t. x ∈ D01

• Solution x01 = (2 3)0 .

• Admissible for both scenarios.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

   min E(F ) = −11x + 3x2

e01 λ 1
Pλ s.t. x ∈ D01

• Solution x01 = (2 3)0 .

• Admissible for both scenarios.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Optimality Test

11π11 6π21


 max +
−2π11 3π21
 !


 s.t. + ≤ 1 π11

−1


−π11 + 2π21 ≤ 0 Sol: =
π21

(P(π1 )) −0.5

 2π11 − 5π21 ≤ 6
π11 6π21

− ≤ 2




π11 π21 ∈

, R

8π12 17π22


 max −
−2π12 3π22
 !


 s.t. + ≤ 5 π12

1


−π12 + 2π22 ≤ 3 Sol: =
π22

(P(π2 )) 0

 2π12 − 5π22 ≤ 2
π12 6π22

− ≤ 1




π12 π22 ∈

, R

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Optimality Test

11π11 6π21


 max +
−2π11 3π21
 !


 s.t. + ≤ 1 π11

−1


−π11 + 2π21 ≤ 0 Sol: =
π21

(P(π1 )) −0.5

 2π11 − 5π21 ≤ 6
π11 6π21

− ≤ 2




π11 π21 ∈

, R

8π12 17π22


 max −
−2π12 3π22
 !


 s.t. + ≤ 5 π12

1


−π12 + 2π22 ≤ 3 Sol: =
π22

(P(π2 )) 0

 2π12 − 5π22 ≤ 2
π12 6π22

− ≤ 1




π12 π22 ∈

, R

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

ä Q (x, sc1) = π1 h1 − T 1 x0 = 2,
 

ä Q (x, sc2) = π2 h2 − T 2 x0 = 4,
 

1 1
ä Q(x) = P1 Q (x, sc1) + P2 Q (x, sc2) = (2) + (4) = 3.
2 2

ä θ = −∞ < Q(x) = 3 ⇒ 2x1 − 5x2 + 4θ ≥ 1

New domain admissibility:

ä
D(02) = D(01) ∩ {x ∈ Nn | 2x1 − 5x2 + 4θ ≥ 1}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

ä Q (x, sc1) = π1 h1 − T 1 x0 = 2,
 

ä Q (x, sc2) = π2 h2 − T 2 x0 = 4,
 

1 1
ä Q(x) = P1 Q (x, sc1) + P2 Q (x, sc2) = (2) + (4) = 3.
2 2

ä θ = −∞ < Q(x) = 3 ⇒ 2x1 − 5x2 + 4θ ≥ 1

New domain admissibility:

ä
D(02) = D(01) ∩ {x ∈ Nn | 2x1 − 5x2 + 4θ ≥ 1}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

ä Q (x, sc1) = π1 h1 − T 1 x0 = 2,
 

ä Q (x, sc2) = π2 h2 − T 2 x0 = 4,
 

1 1
ä Q(x) = P1 Q (x, sc1) + P2 Q (x, sc2) = (2) + (4) = 3.
2 2

ä θ = −∞ < Q(x) = 3 ⇒ 2x1 − 5x2 + 4θ ≥ 1

New domain admissibility:

ä
D(02) = D(01) ∩ {x ∈ Nn | 2x1 − 5x2 + 4θ ≥ 1}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

ä Q (x, sc1) = π1 h1 − T 1 x0 = 2,
 

ä Q (x, sc2) = π2 h2 − T 2 x0 = 4,
 

1 1
ä Q(x) = P1 Q (x, sc1) + P2 Q (x, sc2) = (2) + (4) = 3.
2 2

ä θ = −∞ < Q(x) = 3 ⇒ 2x1 − 5x2 + 4θ ≥ 1

New domain admissibility:

ä
D(02) = D(01) ∩ {x ∈ Nn | 2x1 − 5x2 + 4θ ≥ 1}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

ä Q (x, sc1) = π1 h1 − T 1 x0 = 2,
 

ä Q (x, sc2) = π2 h2 − T 2 x0 = 4,
 

1 1
ä Q(x) = P1 Q (x, sc1) + P2 Q (x, sc2) = (2) + (4) = 3.
2 2

ä θ = −∞ < Q(x) = 3 ⇒ 2x1 − 5x2 + 4θ ≥ 1

New domain admissibility:

ä
D(02) = D(01) ∩ {x ∈ Nn | 2x1 − 5x2 + 4θ ≥ 1}

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

   min E(F ) = −11x + 3x2

e02 λ 1
Pλ s.t. x ∈ D02

• Solution x01 = (2 3)0 e 01 = (0 7 4)t , θ = 3

F
• Efficient

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

   min E(F ) = −11x + 3x2

e02 λ 1
Pλ s.t. x ∈ D02

• Solution x01 = (2 3)0 e 01 = (0 7 4)t , θ = 3

F
• Efficient

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

   min E(F ) = −11x + 3x2

e02 λ 1
Pλ s.t. x ∈ D02

• Solution x01 = (2 3)0 e 01 = (0 7 4)t , θ = 3

F
• Efficient

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

• initial list: L1 ←− {F
e 01 }.

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of domain
The following constraints remove the solution x01 end other so-
lutions.
• F ek (x01 ) − 1)y`i + Mk (1 − y`k )
ek (x) ≤ (F

• y`k ∈ {0, 1}
p
X
• y`k ≥ 1
k=1

−3x1 + x2 + 9y11 ≤ 5, 5x1 − 2x1 + 8y12 ≤ 11,

 
 
1
 
 2x 1 − x2 + 4y3 ≤ 4

 

 
1

D1 = D01 ∩ yk ≤ 1, k = 1, 2, 3,
 X3 
 
y1k ≥ 1, yk ∈ N

 


 

k=1

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Reduction of domain
The following constraints remove the solution x01 end other so-
lutions.
• F ek (x01 ) − 1)y`i + Mk (1 − y`k )
ek (x) ≤ (F

• y`k ∈ {0, 1}
p
X
• y`k ≥ 1
k=1

−3x1 + x2 + 9y11 ≤ 5, 5x1 − 2x1 + 8y12 ≤ 11,

 
 
1
 
 2x 1 − x2 + 4y3 ≤ 4

 

 
1

D1 = D01 ∩ yk ≤ 1, k = 1, 2, 3,
 X3 
 
y1k ≥ 1, yk ∈ N

 


 

k=1

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion


min E(Fλ ) = −11x1 + 3x2

e02
Pλ s.t. x ∈ D1

• Sol. opt. x1 = (1 3)0 → F(x

e 1 ) = (3.5 2.5 2.5)t , θ = 3.5

• Efficient

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion


min E(Fλ ) = −11x1 + 3x2

e02
Pλ s.t. x ∈ D1

• Sol. opt. x1 = (1 3)0 → F(x

e 1 ) = (3.5 2.5 2.5)t , θ = 3.5

• Efficient

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion


min E(Fλ ) = −11x1 + 3x2

e02
Pλ s.t. x ∈ D1

• Sol. opt. x1 = (1 3)0 → F(x

e 1 ) = (3.5 2.5 2.5)t , θ = 3.5

• Efficient

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

The following table summarizes all different steps

Iteration x` θ F
e L
1 (2 3)0 3 (0 7 4)0 L1 = {eF1 }
2 (1 3)0 3.5 (3.5 2.5 3.5)0 F1 , e
L2 = {e F2 }
3 (1 4)0 4.75 (5.75 1.75 2.75)0 L3 = L2 ∪ {e F3 }
4 (0 4)0 5.25 (9.25 − 2.75 1.25)0 L4 = L3 ∪ {e F4 }
5 (0 5)0 6.5 (11.5 − 3.5 1.5)0 L5 = L4 ∪ {e F5 }

Table: The results obtained at each iteration

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

The following table summarizes all different steps

Iteration x` θ F
e L
1 (2 3)0 3 (0 7 4)0 L1 = {eF1 }
2 (1 3)0 3.5 (3.5 2.5 3.5)0 F1 , e
L2 = {e F2 }
3 (1 4)0 4.75 (5.75 1.75 2.75)0 L3 = L2 ∪ {e F3 }
4 (0 4)0 5.25 (9.25 − 2.75 1.25)0 L4 = L3 ∪ {e F4 }
5 (0 5)0 6.5 (11.5 − 3.5 1.5)0 L5 = L4 ∪ {e F5 }

Table: The results obtained at each iteration

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

F inal L ist of E fficient S olutions

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

F inal L ist of E fficient S olutions

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

F inal L ist of E fficient S olutions

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

F inal L ist of E fficient S olutions

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

F inal L ist of E fficient S olutions

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

F inal L ist of E fficient S olutions

e 01 ), F(x
EFFS={F(x e 1 ), F(x
e 2 ), F(x
e 03 ), F(x
e 4 )}
Djamal CHAABANE & Fatma MEBREK
 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

Plan

1 Introduction
The Problem
Motivations
Passage to deterministic equivalent problem of MOSILP
2 Theoretical Developments
3 Principal Components of the Technique
4 The algorithm
5 Illustration
6 Conclusion

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

C onclusion

• Improving by avoiding test-efficiency problem

• Experiment study is very complicated, but possible

• Resolution when the main criterion is not linear

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

C onclusion

• Improving by avoiding test-efficiency problem

• Experiment study is very complicated, but possible

• Resolution when the main criterion is not linear

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

C onclusion

• Improving by avoiding test-efficiency problem

• Experiment study is very complicated, but possible

• Resolution when the main criterion is not linear

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
A

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
Y

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
F

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T
T

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T
T
E

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T
T
E
N

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T
T
E
N
T

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T
T
E
N
T
I

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T
T
E
N
T
I
O

-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

6T H
AN
K
YO
U
FO
R
Y
O
U
R
A
T
T
E
N
T
I
O
N
-
return

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For all possible realizations of ξ

max ut h(ξ r ) − T(ξ r )x0





s.t. ut W ≤ 0

(FDual )

 kuk1 ≤ 1
u ∈ R`

continu

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

For all possible realizations of ξ

 max π t (h(ξ r ) − T(ξ r )x)


continu

(Ftest ) s.t. π t W ≤ (qr )t

π ∈ R`


Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

T heorem
 K
P



 max Θ(ψ1 , . . . , ψK , x1 , . . . , xn ) = ψi
 i=1
n n

 
 X X
ecij xj + ψi = ecij xj0

∀i ∈ {1, . . . , K}
 


 

0



P x 

 j=1 j=1


 s.t. x = (x1 , . . . , xn ) ∈ D
ψi : are real nonnegative integer variables for all i

 

 

ecij : is the jth component of row vector

 


 

eci in problem (P02 )
 

continu

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion

The reduced problem


max φ(x) = E(d)x
(PR )
s.t. x ∈ D`

continu

Djamal CHAABANE & Fatma MEBREK

 Introduction
Theoretical Developments
Principal Components of the Technique
The algorithm
Illustration
Conclusion


 “ min ”F
eλ = (E(F)) + θ
λ
(Pλ ) s.t. x∈D
θ ≥ Q(x)


retour

Djamal CHAABANE & Fatma MEBREK