Distributed Primal-Dual Saddle Point Optimization Applied to Elevator Group Control Systems
Abstract
Elevator group control presents an opportunity for the application of distributed optimization algorithms. A distributed approach to elevator group control could make the system more robust to component failure, as the unaffected elevators can continue operation when one of the controllers fails to function. In this paper, we present a novel algorithm based on distributed primal-dual saddle point optimization. This algorithm is benchmarked against the ant colony optimization algorithm, which is well estalished in the literature, as well as two simple dispatching algorithms. All four algorithms were implemented and simulated in a self-developed elevator group control framework using models of people flows in buildings which are well established in the literature. The developed distributed controller outperforms the benchmark controllers for some building sizes and elevator systems.
This work was part of the course “Advanced Topics in Control” taught by Prof Dörfler at ETH Zurich.
Group work with Jeroen Buitendijk.
Download the full report here.