Mon.2 13:45–15:00 | H 3013 | MUL
.

Applications of Multi-Objective Optimization (2/2)

Chair: Alberto De Marchi Organizers: Alexandra Schwartz, Gabriele Eichfelder
13:45

Floriane Mefo Kue

joint work with Thorsten Raasch

Semismooth Newton methods in discrete TV and TGV regularization

We consider the numerical minimization of Tikhonov-type functionals with discrete total generalized variation (TGV) penalties. The minimization of the Tikhonov functional can be interpreted as a bilevel optimization problem. The upper level problem is unconstrained, and the lower level problem is a mixed l1-TV separation problem. By exploiting the structure of the lower level problem, we can reformulate the first-order optimality conditions as an equivalent piecewise smooth system of equations which can then be treated by semismooth Newton methods.

14:10

Zahra Forouzandeh

joint work with Maria do Rosário de Pinho, Mostafa Shamsi

Multi-objective optimal control problems: A new method based on the Pascoletti-Serafini scalarization method

A common approach to determine efficient solutions of a Multi-Objective Optimization Problems is to reformulate it as a parameter dependent Single-Objective Optimization: this is the scalarization method. Here, we consider the well-known Pascoletti-Serafini Scalarization approach to solve the nonconvex Multi-Objective Optimal Control Problems (MOOCPs). We show that by restricting the parameter sets of the Pascoletti-Serafini Scalarization method, we overcome some of the difficulties. We illustrate the efficiency of the method for two MOOCPs comprising both two and three different costs.

14:35

Alberto De Marchi

joint work with Matthias Gerdts

Mixed-Integer Linear-Quadratic Optimal Control with Switching Costs

We discuss the linear-quadratic optimal control of continuous-time dynamical systems via continuous and discrete-valued control variables, also considering state-independent switching costs. This is interpreted as a bilevel problem, involving switching times optimization, for a given mode sequence, and continuous optimal control. Switching costs are formulated in terms of cardinality, via the L0 ''norm''. An efficient routine for the simplex-constrained L0 proximal operator is devised and adopted to solve some numerical examples through an accelerated proximal gradient method.