Global optimization methods for black box problems
In many situations it is necessary to optimize functions whose analytical expression is unavailable and whose evaluation requires huge CPU times.
An example in this context is parameter calibration in simulation models, where, given some real observed data, it is required to find simulation parameters which produce, through simulation, data which is as close as possible to real observations.
A very important class of global optimization algorithms in this field is that based on surrogate functions, where interpolation or regresion is used to replace the function to be optimized with a simples one; based on this function the next evaluation point for the objective function is decided.
The thesis concerns studying and implementing some variations of methods already included il our lab’s library.
Supervisors: Fabio Schoen, Andrea Cassioli
