Journées de l'optimisation 2009 Optimizations days

NOMAD is a black-box optimization software. The functions representing the objective and constraints of such problems have no exploitable structure (such as derivatives). They are typically evaluated by computer codes. The presentation gives an overview of the Mesh Adaptive Direct Search algorithm and describes the NOMAD implementation and its use.

Hydroelectric reservoirs management requires precise estimation of snow quantity, that becomes a natural inflow in spring. Snow quantity are measured at fixed sites by GMON devices. Krieging interpolation estimates the quantity of snow over the territory and provides an error approximation. We use the recent MADS algorithm to identify the optimal GMON positioning that minimizes the prediction error.

Identification of good algorithmic parameters may be formulated as a black-box optimization problem, and solved using a direct search algorithm. The present work refines this idea to increase the flexibility and emphasize on the black box separation into two components representing the data and structure. We apply it to the derivative-free algorithm of Conn-Scheinberg-Toint.