Journées de l'optimisation 2022

HEC Montréal, Québec, Canada, 16 — 18 mai 2022

Horaire Auteurs Mon horaire

MB8 - Derivative-free and Blackbox Optimization II

16 mai 2022 15h30 – 17h10

Salle: METRO INC. (jaune)

Présidée par Sebastien Le Digabel

3 présentations

  • 15h30 - 15h55

    Handling binary, unrelaxable and hidden contraints in blackbox optimization

    • Charles Audet, GERAD - Polytechnique Montréal
    • Gilles Caporossi, GERAD, HEC Montréal
    • Stéphane Jacquet, prés., Polytechnique Montréal

    In blackbox optimization, some constraints are not given or unknown to the user: they are called hidden constraints. This can lead to evaluations of the blackbox failing to complete and returning outputs such as NaN or inf. The lack of information appears also with unrelaxable constraints, which gives unreliable outputs if violated, and binary constraints. This work uses the few information from those constraints to sort list of points that are suggested by an algorithm using the opportunistic strategy, like MADS. Numerical results are performed on its computational version Nomad.

  • 15h55 - 16h20

    Quantifying uncertainty with ensembles of surrogates for blackbox optimization

    • Renaud Saltet, prés., Polytechnique Montréal
    • Charles Audet, GERAD - Polytechnique Montréal
    • Sébastien Le Digabel, GERAD, Polytechnique Montréal

    This work introduces an extension to ensembles of surrogates, enabling them to quantify the uncertainty on the predictions they produce. The resulting extended ensembles of surrogates behave as stochastic models and allow the use of efficient Bayesian optimization tools. The method is incorporated in the search step of the mesh adaptive direct search (MADS) algorithm to improve the exploration of the search space. Computational experiments are conducted on seven analytical problems and four engineering problems. The results show that the proposed approach improves the performance of the MADS algorithm on complex engineering problems.

  • 16h20 - 16h45

    SOLAR: A solar thermal power plant simulator for blackbox optimization benchmarking

    • Sébastien Le Digabel, prés., GERAD, Polytechnique Montréal
    • Charles Audet, GERAD - Polytechnique Montréal
    • Miguel Diago, GERAD
    • Christophe Tribes, Polytechnique Montréal
    • Viviane Rochon Montplaisir, GERAD
    • Bastien Talgorn, Polytechnique Montréal
    • Aimen Gheribi, École Polytechnique de Montréal
    • Mathieu Lemyre Garneau, Polytechnique Montréal

    This work introduces SOLAR, a collection of optimization problems provided as a benchmarking tool for blackbox solvers. Each problem optimizes the design of a concentrated solar power plant defined as a blackbox numerical model. The type of variables, dimensionality, and number and type of constraints are different across problems. Optimization may be single or biobjective. The solar plant model considers several subsystems: a heliostats field, a central cavity receiver, a molten salt thermal energy storage, a steam generator and an idealized power block.