10 juillet 2003
Artelys participe au symposium ISMP
2003
Artelys participe au 18th
International Symposium on Mathematical Programming (ISMP
2003) qui se tiendra à Copenhague du 18 au 22
août 2003. Artelys présentera des travaux de
recherche liés aux composants Artelys
Dualis et Artelys DynOpt.
A bundle method for convex optimization:
implementation and illustrations
Coralie Triadou (Artelys / CERMSEM)
co-authors :
Claude Lemaréchal (INRIA)
Jean Maeght (Artelys)
Arnaud Renaud (Artelys)
We present the implementation of a bundle method, oriented
toward Lagrangian relaxation of combinatorial problems. Problems
to be solved have a non-differentiable objective function,
and bound constraints on the variables. The bundle method
is a cutting-plane method, in which the deviation of the next
iterate from the reference point is penalized by a quadratic
term. The next iterate is given by a quadratic programm, ours
is a two level one in order to take into account the bounds
on the variables. A dynamic and a static management of the
penalization's coefficient will be compared. The presented
tests belong to different categories : L1- and L2-norm
minimization, Lagrangian relaxations of combinatorial problems,
such as traveling salesman (Held & Karp) or p-median.
We present results for our bundle method Artelys
Dualis on these problems, outlining the peculiarity of
each category.
Modeling, Optimization and Simulation of Stochastic Dynamic
Systems
Jean-Pierre Goux (Artelys)
co-authors :
Nicolas Bonnard (Artelys)
Olivier Teytaud (Artelys)
Arnaud Renaud (Artelys)
The management of hydro reservoirs, natural gas storages,
financial assets portfolios or cement reserves requires to
make a sequence of decisions balancing present gains and expected
future gains. Complex dynamic management strategies that satisfy
demand while maximimizing expected profit and minimizing risk
are necessary. Artelys DynOpt
is a C++ model building and solving environment that facilitates
the description and the analysis of such systems. Artelys
DynOpt implements powerful stochastic dynamic programming
algorithms to compute robust management strategies and Monte-Carlo
simulations for risk analysis. Design and algorithmic principles
will be addressed. Applications in the area of energy stock
management, financial contract pricing and supply chain management
will be presented.
Retour |
