Dp 463 en
Distributed predictive control
Author: Endel Petr
In this thesis the framework of decomposition methods for convex optimization problems is investigated and applied, using the subgradient methods for solving the master algorithm. Potentially large engineering problems - distributed model predictive control and industrial energy network optimization - are solved in a distributed way. Projected subgradient methods are applied to solve master problem of dual decomposition algorithm to obtain optimal solution of the problems. The comparison to centralized approach is provided, pointing out the main features of the distributed approach. Different step size rules of subgradient methods (together with advanced rules by Polyak and Nesterov with their modifications) are implemented, their convergence properties are compared and their implementation demands are also assessed. In the energy network optimization the enhanced method introducing so called net constraints is successfully applied and the influence of electricity price is investigated.
- Petr Endel, tel:+420 777 598 736, mailto: firstname.lastname@example.org