Solución de un problema de asimilación de datos y un problema de localización óptima mediante métodos de optimización binivel

Abstract

Data assimilation problems have been widely studied in numerical weather prediction as a technique for the reconstruction of the atmosphere’s initial condition. Taking this problem as motivation, our goal in this project is finding the optimal placements of the locations of a data assimilation problem, represented by a parabolic equation. We worked in a bilevel optimization problem where the inner level solves the data assimilation problem, and the upper level solves the optimal placement problem. To solve the data assimilation problem, we used a variational approach 4D􀀀VAR. The existence and uniqueness of the data assimilation problem were demonstrated using usual techniques in optimization. By using the Lagrangian approach, we derive the optimality system of the inner problem. As result, we got an adjoint equation with measures on the right-hand side. This was due to the structure of the objective functional. We proved the existence of a unique very weak solution of the adjoint equation by using the transposition method. The numerical solution was also worked in two levels. The inner problem was solved by using the BFGS method while the upper level used the BFGS projected method through an estimation of active sets.