Title: Blackbox optimization with hidden constraints

Speaker: Delphine Sinoquet
(IFP Energies Nouvelles)

Summary:
Robust design of complex systems, such as wind turbines, often leads to optimization problems involving time-consuming simulators. These simulators may encounter failures or instabilities for certain input sets, such as convergence issues in the numerical schemes of partial differential equations. The inputs that trigger these failures are usually unknown in advance and represent a hidden constraint, often referred to as a crash constraint. Since observing a simulation failure can be as costly as a successful simulation, our goal is to identify the feasible region to steer the optimization process away from areas prone to simulation failures.
To address this, we propose a methodology for derivative-free optimization that integrates hidden constraints identified through a Gaussian Process Classifier. Additionally, we introduce an enhanced strategy that focuses on refining the classifier based on an enrichment criterion, when it is useful for optimization convergence. We will demonstrate our methodology coupled with different optimization algorithms using both toy problems and a practical application in wind turbine design.

Biography:
Delphine Sinoquet is a research engineer at IFP Energies Nouvelles (IFPEN), French public research institute in the field of energy, mobility and the environment. She is currently leading a research project in the field of optimization and control of complex systems and supervised 13 PhD students with applications in various fields such as wind energy, geosciences, electrical mobility and environmental monitoring. She coordinated and participated in collaborative research and industrial projects such as recently SAMOURAI project funded by the French National Research Agency (ANR) and CIROQUO consortium dedicated to robust optimization and uncertainty quantification.

WED 06 NOVEMBER 2024 18:00-19:00 (AEDT, Melbourne Time/ 08:00 CET)
ZOOM MEETING ID: 873 1557 5255; PASSWORD: 778635